Hi Joerg,
Joerg said:
Hi Ken,
That is one of the less understood phenomena in electronics. Again
something that isn't usually taught at the university, at least not in
my days. Most people don't realize that nearly all that 1A has to come
out of a tiny capacitor or two. The output cap is the other issue. You
and I and most of the NG participants know that ripple current specs are
something to be reckoned with but many engineers out there don't.
I soon learned that when designing AC drives - it turns out that capacitor
ripple current is really the key to long-term reliability (and cost). Drive
lifetime is governed almost entirely by capacitor lifetime (provided of
course both the design and the production system work well), which is a
function of internal cap temperature - and it doesnt matter from whence the
temeprature arose - ambient, hot devices nearby or Resr*I^2 (or Vcap*Ileak)
loss inside the cap. I have carefully peeled open the tops of many an
electrolytic, and stuck thermocouples down inside them to measure core
temperature directly.
When you buy $1,000,000 worth of caps from Hitachi they are very nice, and
will make caps with embedded thermocouples
They also give you LOTS of
data that joe public never sees - like for example the distribution, mean
and std deviation of measured capacitor lifetime etc. IIRC Hitachi
use -3s.d. as their "rated" lifetime - that way they have a high probability
of every device meeting the specs....
back then (late 80's/early 90's) few manufacturers gave you detailed info on
lifetime. United (Nippon) Chemicon are GREAT - everyone should read this:
http://www.chemi-con.com/u7002/life.php
Its a great description of exactly how to do it. I have also learned (the
hard way) to stick with reputable manufacturers - cheaper is generally much
worse, regardless of what the datasheet says. A common sneaky trick is to
set the end of life spec at -15% capacitance and 300% leakage current, where
most manufacturers use -10% and +200% respectively. This makes "end of life"
quite a bit bigger than it really is. I havent looked for a few years, but
previously every time I did all of the "good" manufacturers were actually
quite similar, and every "new" manufacturers (usually korean) part that my
suppliers found that looked quite a bit better (say +30% lifetime) turned
out to be some such specmanship, and invariably they were quite a lot worse
than my preferred sources.
some manufacturers now make large electrolytic cans with a big hole in the
centre - 20mm or so - to greatly improve the cooling. Removing the inner
portion of a short, fat cylinder makes a small reduction in capacitance, but
drastically improves the thermal behaviour, so such a cap can take a higher
ripple current for the same lifetime.
We used to poke the large cans thru the heatsink and expose them to external
air. I hit upon the interesting yet obvious idea of not putting them in the
middle of the heatsink ie not blowing nice hot air over them. This allowed
us to drop the number of caps per sub-assembly from 8 to 6. In a big drive
we would thus save around 16 caps, and they were about $30 each.
With respect to drivers in these chips many of them provide a push-pull
stage with 10-20 ohms. When I read that this was what convinced me to
swallow the pride and use a chip for the first time. Well, that and the
90 cent price tag of the 3478.
Regards, Joerg
I reckon another area many engineers overlook is resistor transient power
dissipation. Most just look at the average power dissipation. I have seen
many 0603 4R7 gate drive resistors, running from 12V supplies. By the time
Cg' is such that Rg*Cg' > gate driver Tr (Tf), Rg sees about (12^2)/4.7 =
30W peak! And if you toast Rg its value invariably rises, leading to
destruction of the fet which usually shorts D-G and totally destroys Rg....
Regards,
Terry
