Life Test - MMC cap bank

["Tesla list" <tesla-at-pupman-dot-com>]

Original poster: "Terry Fritz" <teslalist-at-qwest-dot-net>

Hi Matt,

It's not that bad.  NIST has all the data for such things:

http://www.itl.nist.gov/div898/handbook/

In our case, the life is a wear out (as opposed to random failure).  The 
graph will look like this.  Probably a pretty high gamma case.

http://www.itl.nist.gov/div898/handbook/apr/section1/gifs/flifepdf.gif

The mean and variance can all be figured out:

http://www.itl.nist.gov/div898/handbook/apr/section1/apr166.htm#Derivation%20and%20Use%20of%20the%20Birnbaum-Saunders%20Model:

But really, it is not that complex.  If the caps last 20 minutes, we 
worry.  If they last 20 days, we don't...  A pretty simple test set up 
would look like:

http://hot-streamer-dot-com/temp/CapLifeTest.gif

5 caps all failing pretty close to each other in time will have a perfectly 
reasonable confidence for our purposes.  Unlike light bulbs, these caps 
seem free from "defects" that cause very early failures and other random 
things to cloud the data.  They should fail like clockwork...

One could also set up a bunch of those 10Meg 1/2W bleeder resistors 
too.  People always worry about those and the voltage too.

I have a Agilent 34070a data logger and 20 channel card coming late next 
week.  This might be a good excuse to use it ;-))

Cheers,

         Terry


At 12:48 AM 7/5/2003 -0400, you wrote:
>In a message dated 7/4/03 4:54:43 PM Eastern Daylight Time, 
>tesla-at-pupman-dot-com writes:
>
>
>>One has to be careful since it may
>>burn one's house down if they flame out when your not there and there is a
>>big electrical hazard and all that.  But that test would pretty well define
>>the lifetime.  It is interesting to note that at first they will just
>>self-heal.  But eventually, they will obviously go "bad"
>
>
>Actually, it would take dozens (or hundreds) of such tests with identical 
>MMC configuration to come up with a useable "mortality table" to determine 
>a "well-defined" lifespan. Statistics 101: "A single observation 
>establishes almost nothing. As a best guess, you can say that the 
>observation is probably between the 20th and 80th percentile."
>
>Matt D.