One of the earlier posters suggested that there is some kind of leakage
effect that pumps the junction between the capacitors up to a high dc level,
high enough to prevent the AC voltage from causing a reverse bias.
Given a diode over one of the capacitors pointing towards the junction, I
can see that it would get pumped up to the highest voltage, minus the .6
drop on the diode. If the diode was pointed away from the junction, then the
opposite would be true, ie, the junction would drop to the lowest voltage +
the .6 PN drop.
If there is asymetrical leakage of current, then this would also happen
without the diode, one way or another. If the asymetric leakage is toward
the junction, then the junction would pump up to the highest voltage. If its
away from the junction, then the junction would drop to the lowest voltage.
Given the above mentioned posting, my guess was that it will leak current
from - to + more readily than + to -. Thus, I put on goggles, and tried the
experiment again, this time joining the + leads rather than the - leads.
Also, I used caps with a higher voltage (50 VDC) rating. The maximum voltage
pp at any one time is about 40v, so I guessed that it should be better able
to handle the stress.
The first experiment was with a diode from the - to + terminal of one of the
caps. The DC voltage between the caps grew to 16vdc, as expected.
Next, I discharged the caps, and tried it again without the diode. The caps
did not explode, but the dc voltage seemed to land at a random place
initially, and then slowly climb. Thus, its probably true that the leakage
is greater from - to + than from + to -, at least with these new caps.
However, it seems to take a long time for these 100uf caps to charge up to a
value that will protect them from being charged with incorrect polarity
(without the diode, anyway). I let the caps charge for 30 seconds, and it
looked like they were charging at about 2v per minute. Thus, the 'bad'
polarity charging would occur in my test setup for about 7 minutes, albeit
with smaller and smaller negative voltages all the time.
I did not repeat the experiment with the caps connected in the other
direction, although by symmetry it should simply charge the negative
junction to a large negative value over time. However, I'm still pulling cap
stuffing out of my hair from the last experiment, so I though I'd leave
things as they stand... Any brave souls who want to give it a try might
email me the results at rcmonsen @ comcast.net
I'm guessing that the explosion from before occured because the leakage
current wasn't fast enough to prevent a big negative overvoltage of the
caps, or perhaps one of the caps had a big leak in the wrong direction.
I'd say that in order to replace nonpolarized caps with polarized caps, I'd
use a diode to quickly charge the junction between them up (or down). Point
the diode from the - to the + terminal over one of the caps. Not using the
diode risks an explosion, or at the least degradation of the caps over time
due to reverse voltages for a time while the system charges up.
The other point is that since the caps are in series, the total value is
C1*C1/(C1+C2), so not only do you need two caps, you need twice the value
that the original non-polarized cap had (in addition to the diode). Also,
both of these caps need to have a dc rating at least twice the peak to peak
AC voltage expected.
Regards