[Amps] MTBF Thread...

[sccook1@cox.net]

It would seem to me that the 1000 devices for one hour approach would be great for determining infant mortality, but have little bearing on useful life predictions.

Also, when devices are subjected to steady state life at elevated temperatures (increased temp = increased activation energy = bad things happen at an accelerated rate; e.g., purple plague and Kirkendall voids, metal migration, tin whiskers, et al.) it is actually a fairly benign environment -- at least from a mechanical perspective.  Seems a better approach would be to add constant vibration and liquid to liquid temperature shock while the device is electrically exercised over its full parameter set.  You could also throw in some constant acceleration and mechanical shock, and what the heck... some excursion over worst case temp-altitude, salt fog / spray -- then shoot it with a .45ACP.  For desert, ESD sensitivity testing and viola -- if it makes it through all of that, you can mark it JAN-S!

But you have to be a very evil person to subject hardware to this kind of testing... Oh yeah, and flammability testing, die shear, tensile strength and elongation, Particle Impact Noise Detection, launch shock, and...

-Steve (WG7K)

> 
> From: Steve Katz <stevek at jmr.com>
> Date: 2005/09/14 Wed AM 10:43:05 EDT
> To: 'David Kirkby' <david.kirkby at onetel.net>,  sccook1 at cox.net
> CC: amps at contesting.com
> Subject: RE: [Amps] MTBF Thread...
> 
> ::In response to the thread from David below: If 1,000,000 devices are
> tested for one hour and 100% of the population survives the test, the MTBF =
> 1,000,000 hours by definition.  That is the statistical mean, as a minimum,
> since no devices failed.  If they *all* fail one minute later, the MTBF
> based on the original test is still 1,000,000 hours because the failures can
> be ignored: The test was concluded, and MTBF already calculated.  This is
> exactly and specifically allowed by both the Bellcore-3 standard and
> MIL-HDBK-217 standard for calculating MTBF, and those are the industry
> norms, at least here in the U.S.
> 
> To be fair, normally during MTBF testing a quantity of devices is tested
> until some failures actually do occur, and do not stop after one hour.  This
> yields are far more accurate picture, of course.  However, it would be
> perfectly legitimate to establish the time standard in advance (say, "one
> hour") and test a population, and calculate MTBF based on that.
> 
> -WB2WIK/6
> 
> +++++++++++++++++++++++++++++++++++++++++++++=
> 
> 
> I am however puzzled by the comments of WB2WIK/6 stevek at jmr.com when he 
> states this:
> 
>  > MTBF is a funny statistical tool: Per even
>  > MIL-HDBK-217 and Bellcore standard definitions,
>  > MTBF may be determined in any way that
>  > arithmetically accomplishes the calculaton of a mean.
> 
>  > Including this one: 1 million hours MTBF
>  > may be described as having 1 million
>  > devices in operation for one hour
>  > each, with failure rate of one
>  > device or less.
> 
>  > If 100% of the population all failed
>  > after 61 minutes, this lot would still
>  > have an MTBF of 1 million hours by that definition.
> 
> I can't for the life of me understand how if 100% of devices in a sample 
> of 1,000,000 all failed within 61 mintes can you draw the conclusion 
> that the MTBF is 1,000,000 hours. I would have thought that is 
> statistically valid data to prove the MTBF is <= 61 minutes, rather than 
>   1,000,000 hours.
> 
> Even if 100% work after 60 minutes, the confidence that you could put on 
> any one device lasting 1,000,000 hours is negligsable.
> 
> I have not however read the mil spec - I only looked at the Seagate 
> (might be another disk manufacturer) and determined the MTBF, which 
> might suggest disks last on average for 50-150 years is an incorrect 
> interpretation.
> 
> You know the old proverb - there are lied, damm lies and then there are 
> statistics!
> 
> -- 
> David Kirkby,
> G8WRB
> 
> 
>