[chbot] Is there a statistician in the house?

[Charles Manning]

Anecdote 1:

In the 1980s  one of the power supplies that my employer designed would
overheat within two minutes at full power. After 10 minutes it would fail.
If operated at 70% of full power (typical use) it would last longer and
would fail after about 12 minutes - sometimes catching fire.

The customer was perfectly happy. How can that be?

The PSU was for a missile that had an expected flight time of approx 1
minute. Any lifetime over that was gravy. Indeed, if it wasn't for the PSU
having a minimum mass spec, they would have made the heat sinks smaller so
that it would have failed sooner.

Anecdote 2: NAND flash. Almost all NAND flash chips have some hw errors.
However these are wrapped up in layers of ECC and software to address those
errors. Therefore the errors are handled.

Moral of the story: there's a massive difference between component,
sub-assembly and system failure.



On Mon, Oct 19, 2020 at 1:19 PM Stephen Irons <stephen at irons.nz> wrote:

> We have just had a discussion about whether the test 'finds' or 'causes'
> the failure, and have concluded that the statement was meaningless, because
> there was no specification about expected behaviour.
>
> For example, we want to test that a device input works correctly at its
> maximum specified operating voltage. The test is to apply 220 V to the
> device input. After the test, the input no longer works. Is this a failure?
>
>    - if the input was rated for 12 V, no it is not a failure - the test
>    exceeded the design specification, and we should change the test.
>    - if the input was rated for 500 V, yes it is a failure - the test was
>    within the specification, and we conclude that something is wrong with the
>    unit under test; it might be a design issue or a manufacturing issue
>
> Our response to the client was essentially that: we have found a test that
> makes devices fail in the same way that they are failing in the field.
> However
>
>    - there was no design specification for that parameter
>    - we don't know whether the failing devices are being used within the
>    design specification for that parameter
>    - we don't know whether our test is within the design specification
>    for that parameter
>    - as a result, we don't know whether the problem is a test issue, a
>    design issue or a manufacturing issue
>
> For interest, the parameter is 'shock and vibration'. The test is to load
> a number of known-good units into a vibration chamber (essentially, a
> tumble-drier with the heater turned off), run it for a few hours, and see
> how many still work after that.
>
> The result: 70% failed after 4 hours, and the devices are in a similar
> failed state to failures in the field.
>
> But we have no idea what level of shock and vibration the units are
> experiencing in the field, or whether our test is in the same ballpark as
> that, or whether either of these are close to the (unknown) design limits,
> or whether the manufacturer is using a part that does not meet the design
> criteria.
>
> Stephen Irons
>
> On Sun, Oct 18, 2020 at 22:39, Marshland Engineering <
> marshland at marshland.co.nz> wrote:
>
> But we now have a repeatable test that causes the failure.
>
> I'd find a new test!!! I'm sure you mean 'But we now have a repeatable
> test that finds the failure.' Cheers Wallace
> _______________________________________________ Chchrobotics mailing list
> Chchrobotics at lists.ourshack.com
> https://lists.ourshack.com/mailman/listinfo/chchrobotics Mail Archives:
> http://lists.ourshack.com/pipermail/chchrobotics/ Meetings usually 3rd
> Monday each month. See http://kiwibots.org for venue, directions and
> dates. When replying, please edit your Subject line to reflect new subjects.
>
> _______________________________________________
> Chchrobotics mailing list Chchrobotics at lists.ourshack.com
> https://lists.ourshack.com/mailman/listinfo/chchrobotics
> Mail Archives: http://lists.ourshack.com/pipermail/chchrobotics/
> Meetings usually 3rd Monday each month. See http://kiwibots.org for
> venue, directions and dates.
> When replying, please edit your Subject line to reflect new subjects.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ourshack.com/pipermail/chchrobotics/attachments/20201019/a04dadda/attachment.html>