Main power supply

The perfect unit for the job is the ESP120 from HP. They are dirt cheap and we already have (bought and got donated) 7 of them. The main power supply can deliver about 16.5kW of DC power. (This is at maximum continuous current and limited to 48.3V, that’s 42V+15%.) When not so much power is needed part of the supply can be switched off automatically to conserve power. The idea is to use 6 units, wired in parallel.

The 2950W (max) ESP120 power supply from HP is powerful, rugged and perfect for the job.
The 2950W (max) ESP120 power supply from HP is powerful, rugged and perfect for the job.

How to control the voltage of these units is explained on multiple websites. It can be controlled by a 0-5V analog voltage. The supply also talks back, there is an I2C connection for configuring some settings and for reading out internal measurements. (yay! stats!) Here is some cutted/pasted/edited code that does return some data. (Code and more based on elweb.info data.) We did not yet get the output current out of these units (more people online having the same problem), but there is a shunt on the PCB of the supply. Maybe it’s better to check out if we can use those if the unit does not support current stats over I2C. Or use some external shunts, that might be even simpler.
The latest test shows that we can get the current data out of the units! It’s not that accurate
(+/- 10%), but still very useable! 🙂

^^Further tests show that, sadly, the current stats are available only when the output is 48V or higher.
We still need to measure the current externally.

Paralleling these units is possible as they are configured like that in blade server power supplies. How to do that is not clear yet. (Voltages should be exactly the same for this to work.) But we have the server power supply board that connected them on both sides, so it should be reverse-engineerable!
There is a load balancing connection at one of the control pins, connecting all of the balancing pins together is all it takes! 🙂 Check this scanned article to see how this works. If this is functional with voltages below 48V is not tested yet. But it seems to be an analog system, so probably we’re fine.
We will test this soon though.

The negative output will be earthed, the 3*230V 32A input will be protected by a three phase 30mA earth leakage current protector that is sensitive to DC faults as well. So if one of the supplies fails for some reason, the whole grid power supply is disconnected from the mains completely. Safety third, err, first!

We will put up a schematic of the way we think we should connect them in parallel and how to control them ASAP.