The whole project was fuelled by initally finding really cheap DC-to-PD modules on #AliExpress (links below).
And for cheap but powerful power supplies, there was basically only one logical choice: Used server power supplies.
The #HP#DPS ones are pretty nice and modding-friendly.
JFF, I did a basic #CAD mockup of how the modules and PSUs could fit together.
It was too stupid and cheap to not give it a shot :P
I ordered a handful of either module and tested them thoroughly.
Both support all relevant #usbPD profiles including PPS and a bunch of other fast charge protocols.
Thermally they behave pretty good aswell. Both #thermal pictures were taken after about 15min of 24V input and 60/95W load in open air.
The hottest component of the 100W module is the 5A SMD fuse with over 100°C :D
But our planned thermal solution should improve things a lot ^^
@techbeard milled the two front plates from copper clad FR-4 PCB material.
They feature a bunch of vent slits, so the server PSUs draw in air through/over the PD modules and cool them.
Next, the modules are mechanically assembled into an array.
Four rows of five 65W modules are screwed onto aluminum strips with thermal pads in between.
These strips are then inserted into two 3D-printed side rails, which are then screwed to the front panel.
This is also the thermal solution. Because it should almost never happen that all modules of a row are fully utilized, the heat should spread and thus get carried away by the airflow better.
During soldering of the 65W modules, I noticed that our heat sink concept worked almost too well xD
It was really hard to solder to the power pads of the modules. Even if you heated it with a good soldering iron for over a minute, the other side of the solder joint was barely liquid.
As soon as you took the iron away, the solder solidified instantly.
And the enamel of the copper wire had to burn off too under those conditions, making it even more difficult.
After mounting the 100W modules to an aluminum strip, it was time to test the array.
We didn't have enough 100W capable loads to do a long full-load test, but close enough :D
The thermals look promising. The aluminum strip spreads the heat pretty well and got to around 60°C itself.
The hottest spot on the modules is now the USB port interlock P-FET at 90°C. And that will improve with airflow too.
The 65W modules got tested too.
They stay surprisingly cool. After some time at full load it only reached around 50°C.
And it looks like the thermal solutions works exactly like intended.
As you can see in the thermal video, the modules beside the loaded one heat up noticably too (in comparison to the rows below, which you basically can't see at all, thermally)
During final assembly of the array, the 100W module group suddenly showed a short circuit on the input.
I was puzzled at first, but had a hunch.
And sure enough, the 1mm thermal pad was slightly too thin and the main input ceramic capacitor punched through the thermal pad and shorted out against the aluminum strip.
Luckily it didn't crack ^^"
I fixed it by simply putting another layer of thermal pad on top :D
The housing of the #PDBrick is a piece of rectangular aluminum extrusion.
For mounting the server PSUs I decided to design and 3D-print some plastic rails, so the PSUs can be slotted in.
The DIY PSU backplane will get fixed to the front rails too, so the alignment and mechanical forces of the PSUs being slotted in will be manageable.
To connect two server power supplies in series, the PE reference of the secondary has to be severed for the high-side PSU.
On those HP DPS units, that's pretty easy. You have to open it, file down two metal spacers a bit, put insulating washers between PCB and case and replace the metal with nylon screws.
This only removes the connection between output GND and PE. The metal case remains grounded and the floating output is still referenced to PE through the low-side PSU.
As time began to run out, I quickly threw together the backplane PCB using perfboard, thick copper wire strands and lots of solder. It has to potentially handle up to 100A.
And yes, the wonky speed holes are necessary for improved airflow :P
With the added resistors and bridges, the PSUs turn on and everything works.
Wiring the backplane to the modules was a bit frustrating, because of the thick diameters, thermal mass and tight spacing, but worked out fine.
After a final test of the electronics, everything can be put together. @techbeard drilled and countersunk the necessary holes in the aluminum extrusion and fabricated some corner pieces, so the back plate can be screwed to the extrusion.
We mounted the rails, backplane and #usbPD module array inside the extrusion. Now the PSUs can be slotted in, which is quite satisfying.
The 230V wiring consists of a #powerCON#TRUE1 socket to two IEC plugs. We had to remove the jacket, because it would've been too tight otherwise.
The aluminum casing is grounded via the good connection between PSU casing and aluminum. (see above)
You might also notice the #XT60 plug. It's a generic 24V output for connecting external devices like LiPo chargers etc.
The main reason for its existance is still having "left-over" power budget from the power supplies (~600W) :D
Once it was used for a longer period, we noticed the aluminum extrusion being pretty warm to the touch (~40°C).
Apparently, the server PSUs use around 40W during idle. But when drawing that amount from the output, the power usage doesn't go up that much, so they seem to be burning it internally when having (almost) no load?
Also, maybe the aluminum could be a few degrees colder when painted? Because bare aluminum doesn't seem to radiate much heat.
Or we'll install additional fans.
Radiative heat transfer is driven by surface area and temperature difference vs ambient. But you need to consider convection, too, which is why vertical fins on the side edges might improve cooling by creating air guides. Also, small feet to lift it off the surface for underside convection, unless you plan to sit in on metal or stone surfaces.
Paint would hurt unless it's a very special paint, but anodization wouldn't.
As the server PSUs feature #SMBus connectivity, we planned to display some metadata on an OLED.
I soldered up a small perfboard with a buck converter, #ESP32 and a few connectors.
The ESP32 only communicates with the low-side PSU, as I didn't want to implement isolated I²C.
But I noticed that the values coming from the PSU were pretty hit & miss anyways. Especially during light loads.
So I simply used the values that matched the actual values closest and doubled them.
@LeoDJ I've used that same 65w module for my last minute powerbank project I've cobbled together tor a 14h trainride across Germany:)
It kept my thinkpad alive tor a whole day of coding/compiling
@stfn@LeoDJ you can just look for them on AliExpress (or possibly eBay as well) with something like "65W USB-C fast charge module".
However you should be careful, a lot of laptops need 20V to charge and most of these modules are buck-only (so they can only output 12V or less with 12V input).
There are also buck-boost modules though which can do 20V out with 12V in.
There are some 65W and 100W cigarette lighter socket ones too if you don't want just the module.
Add comment