Asimov and Steve:

I would agree, designs for lower voltage operations (at 50% of spec) tend to be
much more conservative, particularly when the actual cost differential isn't
that great, but it really does seem to cut down the failure rate. Also agree on
the 80% figure, and that's what I'm trying to achieve (at minimum) with the
change to higher voltage caps, (and relayout of the pcb to use 10 caps). Even
if I can't get the higher voltage 500V caps, I can still get nearly 20% margin
at the nominal 3600volt DC level, leaving some extra capacity for an occasional
surge... I remember testing 50 caps at a time in my "blast box" to see how long
they would withstand their surge rating. Yep, more than a few disintegrated
during the testing, and many bulged without actually coming apart, all within as
little as a few hours at the rated surge voltage. If I remember correctly, I
also saw about a 5% failure rating at rated voltage within a few hundred hours,
though the particular units tested at that time weren't computer grade or
inverter grade caps like the part numbers I'm currently looking for. Of course,
testing with a higher ripple content always broke the caps down faster than
nearly pure DC, but you'd expect that just from the heating effect of the ripple
current alone. Sometimes I don't even want to think about the hours that were
dedicated to finding caps with both a low leakage figure that had a decent
margin, and a long life so they could be used on part of a product that spent
about a third of it's time on (lithium) battery power.

Most caps are no longer marked for surge, but the datasheets typically show the
surge rating. Unfortunately, you have to be very careful with surge figures,
because some manufacturers seem to provide "absolute maximum" ratings instead of
safe "momentary" surge ratings. Of course, it's up to the purchaser to determine
whether the published ratings are satisfactory for the intended use (as I was
quoted on many an occasion when inquiring of the manufacturer in years past.
Just try to get a manufacturer to state in writing what they mean by
"momentary"...

Having performed "accelerated aging" tests to help determine the failure rate of
stressed components, and compared it to actual field failures, I know that it
isn't an exact science by any means. At least this is only a capacitor. Just
try to get decent information on a battery (numerous experiments with lithium
thionyl chloride types comes immediately to mind... ) is like searching for
chickens with teeth. If capacitor manufacturers treat their products like some
battery manufacturers did 10 years ago, it's a wonder they even stamp a rating
on them... Of course, a tolerance rating of -10%, +50% on a capacitor should be
a telling item to begin with...
[sigh]

Thanks for the response.
--Rick

Asimov wrote:

"Rick Frazier" bravely wrote to "All" (14 Oct 05 07:08:21)
--- on the heady topic of "High Voltage Caps for Plate Supply ??"

RF From: Rick Frazier
RF Xref: core-easynews rec.radio.amateur.homebrew:88241

RF I have an Ameritron AL-1200 amplifier, and it came (used) with a
RF capacitor problem.
[,,,]
RF I was in high tech engineering (computers and peripherals) for 20
RF years, and we never ran electrolytics at anywhere near their rated
RF voltage. Permissible margins for low voltage DC circuits were 50% or
RF more margin. Typical rule of thumb was 100% margin (a 10 volt cap
RF running on nominal 5 volt line, etc.). Even in an industry where every
RF penny of component cost was significant, submitting a design with
RF electrolytics running with as little 25% margin typically got a less
RF than stellar response during design reviews, to say the least.
[,,,]
RF Thanks
RF --Rick AH7H

Rick, it is okay to run electro's at about 80% of their voltage spec.
Electro's are formed at up to 140% of their final spec and can thus
tolerate turn on surges. It is the turn-on surge spec that one should
worry about because that is when the electro is vulnerable. At turn on
an electro re-forms a little bit which draws a larger current than
usual. If that current doesn't fall off rapidly enough the electro can
become damaged and leak permanently. In the old days electros used to
be stamped with both surge and working voltage ratings.

Isn't using an electro at 50% of spec rather conserative? I might
understand being this prudent with semiconductors. An electro also has
a leakage current spec at the rated working voltage. There isn't a
direct linear relation between voltage and leakage. The relation is
more like a reverse exponential and the leakage current drops off a
lot more at a lower percent of rated voltage.

A*s*i*m*o*v

... Paul's Law: You can't fall off the floor.