Re: kablooey



John Larkin wrote:

On Fri, 10 Mar 2006 16:03:48 GMT, Michael <em.pea.sea@xxxxxxx> wrote:

Just two nights ago a friend who works in the RAID industry, in Failure
Analysis, described a rash of component failures (capacitor punch-through) that
he diagnosed as caused by SMPS spikes at turn-on. 10vdc caps on a 5vdc line
were being killed by 12vdc. Switchers save iron and weight, yes, but poorly
designed ones can be a basket of headache.



Yikes! 12 is even worse than 8.8.

Now I'm beginning to wonder... how many unexplained, seemingly random,
logic failures are caused by smps's and random-duration brownouts?

Many switchers have huge open-loop voltage margins, especially 85..260
volt universal-input types run at 240 volts input. A 5-volt supply,
running wide-open with 240 in, could easily make 15, maybe even 20,
volts out.


Recent posts described a very common startup circuit that has a negative aspect
related to static power dissipation in the high-value res from the rectified line
to the control supply. (I put a FET in there to shut off the R when the supply was
running on mine, to reduce that heat.) Also, if the supply fails to start, that R
can get real hot.

Anyway, another type uses a very flaky "teaser" circuit to blip on of the main
power FETs on every once in a while. If these supplies fail to start for any
reason, they just sit there squeaking and chirping. The idea, when everything
is running right is that the FET is just triggered one time for a few microseconds
every couple of seconds, and just one "tease" is enough to charge up the control
supply and start the whole supply up. This design usually uses fewer parts, and
no big resistor. But, it is a mighty UN-controlled way to start the supply!
I'll almost bet that this is how the supply that started this thread works.
If the variation of the components is too large, or the breakover device that
triggers the FET is flaky and "sputters", it can result in an overvoltage condition.
Even still, the crowbar should be a totally independent function, just a Zener
and an SCR. Not having a crowbar on a multi-thousand $ VME crate that is
guaranteed to have even more thousands of $ of boards plugged into it is totally
unbelievable. Our VME crates run about $7000, and have probably twice that
value of boards in them.

Of course, this may be a case of two points of failure. You'll never know if your
crowbar is working unless you actually test it. So, the crowbar might just have
been non-functional, and nobody noticed until the power supply regulation went
haywire. I can only see one way a momentary power dip would cause an overvolt.
The power supply probably has a slow turn-on arrangement that normally keeps
it stable. The dip was enough that the switching transistors went to 100% (or
full) duty-cycle, thus causing the regulation to go open-loop. Integral windup
occurs, then the mains power comes back on, and the loop can't react fast enough
to rising main supply voltage. Whew, that takes some pretty crude design of
the control loop! If it didn't have slow-up, it would do this EVERY time it was
turned on. Also, a load dump would liekly trigger the same situation. The maker
certainly should test to a worse load-dump than any user could ever cause. Does
this crate allow hot-plugging? If so, then the load dump of a hot-unplugging
needs to be covered.

Jon

.

Relevant Pages

  • Re: kablooey
    ... Just two nights ago a friend who works in the RAID industry, in Failure ... to the control supply. ... power FETs on every once in a while. ... Not having a crowbar on a multi-thousand $ VME crate that is ...
    (sci.electronics.design)
  • Re: kablooey
    ... Just two nights ago a friend who works in the RAID industry, in Failure ... to the control supply. ... power FETs on every once in a while. ... Even still, the crowbar should be a totally independent function, just a Zener ...
    (sci.electronics.design)
  • Re: next i the major series of speeched by cameron....
    ... > he can certainly do nothing without power... ... >>...we will respond to state failure by empowering individuals and civil ... public service as well as material wealth..... ... It reminds me of the Punch cartoon of two rustics meeting, ...
    (uk.politics.misc)
  • Re: Linux 2.6.29
    ... failure for any file system that is built on transactions. ... If you acknowledge back to a client a write, then have a power outage, the client should reasonably be able to expect that the data survived the power outage. ... Well, I can go mainly by my own anecdotal evidence, and so far I've actually had more catastrophic data failure from failed drives than anything else. ... The elevator does not issue write barriers on its own - those write barriers are sent down by the file systems for transaction commits. ...
    (Linux-Kernel)
  • Hardware reliability (was: Re: Linux community software-update-anarchy polemic)
    ... >Because power failures are not the only types of outages. ... *You're arguing using the premise that the surface of a disk is a more ... failure as a justification for that premise. ... Therefore I presume that what you really mean isn't that the ...
    (comp.os.linux.misc)