Hi Steve/Jon,
If you consider the test units as Work In Process (WIP), you
create an mfg inventory location for the test site, and consider this
an extension of the manufacturing process as Post Pack Audit, you
can consider the additional units as new and sell them as such.
Many companies have a post pack audit procedure under
which units are unboxed, tested, reboxed and sold as new after
performance testing is performed. There is nothing wrong with this.
Actually, if I were a customer, I would want one
of the units subjected to the PPA test because I would know it
meets spec.
I do agree that this is impractical for very large units (although
we audited units up to the size of a large refrigerator weighing about
1000 lbs) or very fragile units. It also adds cost for the occasional site
time,
labor, and inventory management. But if it is done right, there is
no good reason that the units need be sold as used instead of
new.
It is obviously also impractical for small companies and low volume
production environments.
WRT to Jon's specific questions, the CISPR 16 formula has the answers
when one thinks about the intent.
1. Xn is the arithmetic mean of a data set. It is meaningless in
the context of the test to lump different frequency harmonics
or sources into the same data set . So
a. The data for the calculation is from at least 3 identically
configured samples of the same model.
b. The data is for a given harmonic from the same source
regardless of margin from the limit. For example, the
5th harmonic of a CPU clock will be considered the same
frequency even though it is not EXACTLY the same due to
variation in part tolerances. The third and 5th harmonic
would not be lumped in the same data set.
2. There is no limit to how many frequencies are evaluated but common
sense usually prevails. For example, a frequency very close to the
limit may have no problem if the standard deviation of the data set
is 0 while a frequency with an Xn 10 dB below the limit may be a
problem
if the sample standard deviation of the data set is 8dB. It is good to
scan
the first unit thoroughly, then spot check a second unit at the same
frequencies. If the data between the first two is very close (say,
within
2dB), then check only frequencies within 6 dB or so from the limit for
all remaining test samples. If the data spread is larger, increase the
margin for the remaining samples to 10dB or whatever the data spread
merits. If you are testing 5 samples, and all frequencies pass by the
test by the time the third sample is scanned, stop there and don't
check the last two. The constant, "k," in the equation takes the
smaller
sample set into account.
Typically, it would take about 16 hours of test time to do a 5 unit
pre-production
audit and as I stated previously, about 30% - 50% of the products I saw had
some
relatively minor problem (usually mechanical) that was corrected due to the
audit. This worked well for high volume production environments where
non-compliant product was shipping by the 1000's of units. It does not
make sense for huge products or low volume as stated above.
I did this for a couple of years so in this case, I comment from experience.
I am not a statistician by any means, but have seen real benefit from such
a program.
One final comment. There is a farily sizable number of ITE products out
there
shipping that do not comply with Class B EMI regs in typical configurations.
I see them all of the time. While I don't believe they
are bad enough to cause major EMI pollution, it makes it very difficult to
requalify this stuff as a VAR in end use configurations. Wouldn't it be
nice
if some of these vendors paid a little more attention to the EMC performance
to these products? Just wishful thinking!...
Ciao,
[email protected]
----------
From: Steve Chin
To: HANS_MELLBERG@non-hp-santaclara; Jon D Curtis
Cc: emc-pstc; Michael_Barge
Subject: Re: Measurement Uncertainty
List-Post: [email protected]
Date: Monday, January 13, 1997 7:27PM
Well put, Jon. Let's not forget that all of that additional testing not only
costs the manufacturer time and test money, but there is the also additional
cost of at least two more sample units which may now no longer be sold as
"new" units (having been used). Some of the units my company manufactures
carry a list price in excess of $50,000 US. My accounting department would
not
like to have me write off that much in equipment every time I test, and my
budget won't support that. In the end, it'd mean that the consumer would end
up paying even higher prices to purchase product from companies who are
trying
to "do the right thing," or that the companies who are conscientious will go
out of business in favor of companies who either don't do testing (yes, they
exist!) and sell lots of stuff to consumers on the cheap, or those that do
minimal testing.
I'd love to be able to take up three OATS and test my equipment. I just
can't
do that. Besides, my test lab doesn't have the site time available to run
all
of those tests for al of their clients.
Steve Chin
StreamLogic Corp.
Menlo Park, CA, USA
[email protected]
The views expressed in this communication belong only to their authors. They
do not necessarily reflect those of their employers or each others'.
--------------------------------------
List-Post: [email protected]
Date: 1/13/97 7:04 PM
To: Steve Chin
From: Jon D Curtis
Dear Hans,
I know of no manufacturers actually engaged in series production audits.
So lets hear from them. Please respond to this forum.
The companies I work with look to CISPR 22 8.2.1.1 and test one sample.
Some of them are happy with 0dB margin. I advise a higher margin, but
they are responsible for signing the DoC. To date it would appear to me
that the 80/80 rule only has a place in making it harder to take product
off the market. You can go to market with only one sample tested, but if
someone wants to restrict your access they have to perform an 80/80 rule
statistical test to say you fail (CISPR 22 8.2.4).
As a test lab, I'd love the 80/80 rule if the market would support it
(three-five times the testing, yippee!). The doctrine also seems to need
a bit of clarification: Xn is refered to as the value of the individual
item. Is this the value of the one point closest to the limit? Can you
change the frequency? On a product do you evaluate more than one
frequency? How many? - the six closest to the limit? When doing more
than one test, are several 80/80 tests performed - one for radiated, one
for conducted? The 80/80 test is a statistician's dream and a test
engineer's nightmare.
Jon D. Curtis, PE
Curtis-Straus LLC [email protected]
One-Stop Laboratory for EMC, Product Safety and Telecom
527 Great Road voice (508) 486-8880
Littleton, MA 01460 fax (508) 486-8828
http://world.std.com/~csweb
On Mon, 13 Jan 1997 [email protected] wrote:
