Doing dangerous things to computer power supplies

[manikyath]

I've been prodding at the idea of making a li-ion based UPS for my home server, that fits in the footprint of an ATX power supply (or rather.. elongated form, i have ~30cm of length to work with in the case)

 

my first idea was the most "obvious":

AC-DC power supply -> input switching circuitry -> wide range input DC-ATX power supply

battery charger -> battery -> also to input switching circuitry

 

the most sensible voltages to go with here is going 7S on the li-ion pack (21-29.4 volts), and have a 32-ish volts AC-DC power supply.

only.. i've found this idea has a combination of downsides:

- all the required electronics are quite costly.. a li-ion charger, an AC-DC power supply, then a DC-ATX power supply...

- i'm targeting about the 750w ballpark for maximum load, so that would mean a current of up to 40A, which means thick conductors, bigger switching logic, ...

- all those conversions dont actually make that efficient of a power supply, running off the mains i'm only expecting to see about 70% efficiency.

 

upside is that 7S is something we have all the stuff for at work, so a BMS and quality charger is easy to come by.

 

-----

 

but then, crazy ideas happened.

 

a PC power supply these days is essentially this:

AC-DC rectifier -> big fat mains voltage cap that everyone on the forum (including me) tells you never to touch -> DC-ATX power supply

 

and sleep-deprived me one morning thought "now, what if i just sent DC voltage into that cap?"

 

the downside here is that a PC power supply expects 90-240 volts on the input side... so the battery pack for this theoretical idea would be at least 30S...

upside would be that i only need to find a 30S BMS (probably just gonna slice the pack in 3 and run 3x 10S BMS'es), a diode array that'll do 10A, and an ATX power supply.

 

what i dont know, however, are the following details:

- i live in 240 volt land, are "wide input" ATX power supplies that claim 90-240 volts *actually* wide input - aka will they handle the voltage in that cap suddenly tanking from 240 to 110-ish?

- is the mains frequency actually used for anything in an ATX power supply? is it as easy as injecting DC in the right place, or does it need the mains "pulse" for something?

 

i'm quite comfortable working on the guts of ATX power supplies, and i can find the necessary expertise for a (stupid) high power li-ion battery pack at work.. it's just the "will this actually come together?" part that i'm not sure of. 

 

if anyone has tried something similar, or has the insight into ATX power supply design to know if ignoring AC all together is an option.. and potentially which brands of power supply are the most "accepting" to varying from 240V to 110-ish, the feedback is highly welcome.

[OhYou_]

I think you will be better off going with your first option, for one small detail.

the BMS for a 34 or more in series pack is abhorrently expensive.  You unfortunately cant safely get away without a BMS if you use lithium. 
You *might* however narrowly get away with it if you use a undervolt/overvolt protection module on each 7s pack which will either shut of charging or shut off the computer. you wont have any balancing but oh well. 

 

You actually do not even need to connect into the mains filter cap on the psu. you can directly connect DC to the AC input in almost all switchmode power supplies. 

it will just pass through the rectifier. 

 

honestly, how I would accomplish this is to take an ordinary PC psu and a wide input dc psu like a picopsu M3 or something, and combine the two 24-pin into one. 

This may sound unsafe but it should be fine. 

next, isolate the -12v rail of the normal pc power supply, you will use this as your grid loss signal. 

now, connect a spare eps or pci-e from the normal PSU to at least a 30A relay, using a normally open config, and the other side of that a beefy cv-cc

use this -12v rail wire on the relay so that it will shut the relay when the regular PSU comes on. 

the cv-cc will be your battery charger, simply set it to your desired battery 100% charge voltage and limit the current to something sensible.

connect the battery pack to that, then connect the battery pack to the picopsu using an appropriately sized fuse. 

 

The two psus will run all the time but they should not fight each other as there is a significant resistance difference between literally wires, and going through the charger and picopsu, so under grid on situations, it will just run completely off of the normal psu. 

once the grid falls, the -12v rail should become unpowered and shut down the charger to prevent a loop, and the pc will just keep going.

This is generally a good idea to verify this works, otherwise you may hit overcurrent on the pico 12v rail as the system will literally be charging itself. 

the system charging itself however, isn't a catastrophic issue to worry about, as you only take the conversion losses at the pico and the cv-cc. you should keep this in mind when choosing a charge current, or the system will abruptly shut down if you have a relay failure to open and lose grid.

 

[mariushm]

A wide range power supply will accept 90v AC or more. 

When you rectify AC voltage, you get a DC voltage with a peak equal to Vdc peak = sqrt(2) x Vac - 2 x forward voltage of diode in rectifier =  1.414 x 90  - ~ 2v  = ~ 125v DC

So, your battery would need to provide at least 125v DC to the power supply.

 

Modern power supplies have an active PFC circuit which take this rectified DC output which varies between 0v and 125v DC peak and tries to boost (raise it up) to around 400v DC - that's the voltage you would have on that big capacitor or capacitors. The controller then sends pulses of this 400v through the main transformer to produce 12v and then 5v and 3.3v.

 

That 125v DC is just peak voltage, not constant voltage... so if the Active PFC circuit is used, it may actually work with even lower than 125v DC, but I don't know, and I'm not sure how much an Active PFC circuit would like to work at that very bottom of operating range.

 

[manikyath]

Just now, OhYou_ said:

the BMS for a 34 or more in series pack is abhorrently expensive.  You unfortunately cant safely get away without a BMS if you use lithium. 

that's why i was pondering slicing the pack into 3 segments with their own BMS, and have some 'essentially a 3S BMS' circuitry on top of that. the BMS'es we have at work do the over/undervoltage protection for each individual pack.

 

4 minutes ago, OhYou_ said:

you can directly connect DC to the AC input in almost all switchmode power supplies. 

it will just pass through the rectifier. 

but the rectifier isnt designed for that (it'll constantly load two of the diodes in the rectifier, instead of pulsing both pairs) - and the rectifier in this is basicly a free "mains backfeed protection" in this matter, too.

 

3 minutes ago, mariushm said:

 

Modern power supplies have an active PFC circuit which take this rectified DC output which varies between 0v and 125v DC peak and tries to boost (raise it up) to around 400v DC - that's the voltage you would have on that big capacitor or capacitors. The controller then sends pulses of this 400v through the main transformer to produce 12v and then 5v and 3.3v.

 

That 125v DC is just peak voltage, not constant voltage... so if the Active PFC circuit is used, it may actually work with even lower than 125v DC, but I don't know, and I'm not sure how much an Active PFC circuit would like to work at that very bottom of operating range.

the ActivePFC doesnt sound like something i'd like to rely on 'from lower than mains voltage' for a constant load, and 125vDC is "equally possible" as 90vDC. (simply, take 3 13s packs, instead of 3 10s packs)

 

any idea if the PFC circuit will have problems interacting with just plain DC? this is something i completely overlooked, actually.

 

[OhYou_]

30 minutes ago, mariushm said:

A wide range power supply will accept 90v AC or more. 

When you rectify AC voltage, you get a DC voltage with a peak equal to Vdc peak = sqrt(2) x Vac - 2 x forward voltage of diode in rectifier =  1.414 x 90  - ~ 2v  = ~ 125v DC

So, your battery would need to provide at least 125v DC to the power supply.

 

Modern power supplies have an active PFC circuit which take this rectified DC output which varies between 0v and 125v DC peak and tries to boost (raise it up) to around 400v DC - that's the voltage you would have on that big capacitor or capacitors. The controller then sends pulses of this 400v through the main transformer to produce 12v and then 5v and 3.3v.

 

That 125v DC is just peak voltage, not constant voltage... so if the Active PFC circuit is used, it may actually work with even lower than 125v DC, but I don't know, and I'm not sure how much an Active PFC circuit would like to work at that very bottom of operating range.

 

actually, the 1.414x increase isnt a factor here. if the ac is 90v RMS, which basically means it is equivalent in power to 90v DC 

where you get the actual increase is in the way the capacitor works, that 90vac rms is actually 125v peak to neutral, and in no load situations, the capacitor will report 125vdc because it is able to hold the voltage high around the peak for the duration of 1/50th or 1/60th of a second. 

It can be ignored in this case by inputting at least 120vDC. 

keep in mind you are increasing load through the rectifier by 1.414x and ideally you should increase your DC voltage by 1.414 or increase your psu capacity by 1.414. 
but its probably gonna have the overhead already.

 

sorry i just realize we are basically talking about the same thing and i made mistakes. what I meant is its 125v peak but under load it will pull down to 90v because it is not able to sustain peak. the 90vdc input would abide by the 1.414x limit to capacity, the 125vdc would not. 

Edited May 12, 2023 by OhYou_

[OhYou_]

7 minutes ago, manikyath said:

but the rectifier isnt designed for that (it'll constantly load two of the diodes in the rectifier, instead of pulsing both pairs) - and the rectifier in this is basicly a free "mains backfeed protection" in this matter, too.

its generally fine, I skipped explaining it but then i did in my next post.

It would provide mains backfeed but you are gonna spend a lot more to match the dc voltage on 230vac than an external mains isolation circuit.