Issues with ATX Power supplies in Serie

[SnipTheCat]

Well the fist thing I did is to break the connection from the ground to the earth reference, so all supplies are floating.

I did try 12V on one supply in serie with 3.3V on another supply, while it did seem to work fine, as soon as I would shutdown the 3.3psu (leaving the 12V on), it would die. 

 It would just not work, no spark, no smoke, and when turning on/off the power switch a faint whisltling with frequency doing down (inspection didn't show any clean component that died, power resistors, FET and diodes are all ok)

Did the same with another supply (same model) an used 12V + 5 to have 17V, worked fine for a while, however the 5V supply stopped to work suddenly  (load going though it without being on would be in the order of 30-40mAmps), It was connected an ups and the ups entered in failure mode. I reconnected it directly to the mains and I got a nice firework from inside the box.

The result is: Fuse burnt: 2 Transistors on the high side burn, couldn't find something else gone wrong.

Any idea if there is an extra protection to add to these cheap supplies in order not to fry them if they "pass through" current while their being turned off?

Edit: Corrected the horrible spelling, typed too fast on smartphone with autocorrect turned off, and while too tired ;-) Sorry.
 

[uncle_bob]

Hi

The simple answer is: Don't do this.

ATX supplies assume things like minimum loads on various outputs. You never see that mentioned in a spec sheet, but the assumption is made.

Even with "commercial" switchers, hooking them in series only works when they are specifically designed to allow this. There generally is a secondary cable that is used and a "master  / slave" sort of arrangement. It is very unusual to find one designed like this. It normally much cheaper to simply do an all in one box solution.

Bob

[Photon939]

ATX PSUs especially older ones are usually group regulated, which means all the voltages come from taps on the main transformer with only one feedback loop. Loading only one voltage output can cause the others to go way out of spec and trigger the overvoltage protection circuit.

[SnipTheCat]

I have read about this, that the 5V needs to be loaded in order to have a correct regulation, the supply that was providing 12V has a resistor load on the 5V rail (4 ohm / 50W resistor), I didn't load the 3.3 nor 12V on the one the supplied the 5V.

While I admit it's not the best idea to do so (had plenty of ATX supplies around in a box and needed 15A+ on 16-18V with no suitable supply available around), I just wanted to know the reason behind the failure. Now, anyway... seen the inside of those supplies I wouldn't dare to use them on a computer they are quite horrible... I did recover some pieces (mostly the heatsinks which are decent for other uses, fan, an optocoupler and dual diodes) the remains went to the trash.

[macboy]

First, you must ensure that you have a reverse biased diode (Schottky preferable for low V_F) across each rail. This provides a current path in the event that one of the supplies is trying to push current and the other isn't, such as during power-up, power-down, and maybe during an overcurrent protection event. These diodes are essential. Without them you will damage one or both supplies. The diode must be able to pass the full current you will expect, so perhaps as much as 30 A or more. A big TO-3p or TO-247 rectifier salvaged from a secondary of an ATX supply works well.

Second, ensure that you have a minimum load on each supply, at least 500 mA on the 12 V rail. You usually don't need to load all rails since they are usually derived from the same transformer and are regulated together, though with high-end boutique supplies, this might not be the case. 

When you defeated the ground, I hope that you left the ground connected on the primary side, but only disconnected it from the secondary side. The metal enclosure and several essential components on the primary side must be grounded. If all this is not the case, fix it. Usually the secondary is ground-referenced by a connection to the metal case at one of the PCB mounting screws, so floating the secondary is as easy as isolating this connection. Of course you do so at your own risk; although the secondary is transformer isolated, it was also designed and built with a connection to ground.

[SnipTheCat]

First, you must ensure that you have a reverse biased diode (Schottky preferable for low V_F) across each rail. This provides a current path in the event that one of the supplies is trying to push current and the other isn't, such as during power-up, power-down, and maybe during an overcurrent protection event. These diodes are essential. Without them you will damage one or both supplies. The diode must be able to pass the full current you will expect, so perhaps as much as 30 A or more. A big TO-3p or TO-247 rectifier salvaged from a secondary of an ATX supply works well.

Second, ensure that you have a minimum load on each supply, at least 500 mA on the 12 V rail. You usually don't need to load all rails since they are usually derived from the same transformer and are regulated together, though with high-end boutique supplies, this might not be the case. 

When you defeated the ground, I hope that you left the ground connected on the primary side, but only disconnected it from the secondary side. The metal enclosure and several essential components on the primary side must be grounded. If all this is not the case, fix it. Usually the secondary is ground-referenced by a connection to the metal case at one of the PCB mounting screws, so floating the secondary is as easy as isolating this connection. Of course you do so at your own risk; although the secondary is transformer isolated, it was also designed and built with a connection to ground.

Thanks for the infos ;-) I did indeed isolate the supply from the ground by isolating the mounting screws, the case is still grounded, but I don't see any ground connection to the primary (or is though the case & mounting screw as well?)

Of the 2 supplies that died, for one I couldn't find what's wrong (no burnt component, it's just not working while not drawing anything on the 220V AC)... for the second one it's the fuse + 2 of the power transistors of the primary that blew up (which seems rather odd... I would have expected the secondary to blow)

I did recover dual schottky diodes from the failed supplies they are rated for 2x15A, however I think I'll just wait for a proper 18V/30A supply (A little complicated in Thailand)

[mav_iqdirect]

Consider this schematic. If you have 3.3V supply switched off what happens? For sure supplies have internal resistance, filter caps inside. Maybe also reverse diodes. So we have a real chance of 12V applied in reverse to our 3.3V supply burning it out.

Computer power supplies behave badly if connected in series. Smarter ones just shut off and cheaper ones break down.

[macboy]

Consider this schematic. If you have 3.3V supply switched off what happens? For sure supplies have internal resistance, filter caps inside. Maybe also reverse diodes. So we have a real chance of 12V applied in reverse to our 3.3V supply burning it out.

Computer power supplies behave badly if connected in series. Smarter ones just shut off and cheaper ones break down.

This is precisely why you need to add the reverse-biased diode across both supplies. If one shuts off for any reason, the diode conducts, ensuring less than 1 V of reverse voltage across the disabled supply. All lab supplies have this diode for this exact reason.