I just went through the process of building a DCDC UPS system. It is meant to power some network essentials here for about 24hrs or so. By virtue of being a DCDC design, it skips the usual AC conversion step and I expect a much higher overall efficiency.
Question is: are any of you interested in hearing about this quest and ready to have a laugh at it or is it of no interest?
I’m interested 
Likewise. UPSs in general have been a very frustrating part of the market.
Cool. The design is supposed to provide power to a BlueIris system for hours, possibly days. Key is avoiding AC-DC conversions to improve energy efficiency and hence maximize runtime. So what has to run and how do we run it?
As a starting step, I would take stock of what you have and what you actually need to run - first in terms of equipment, but also in terms of voltages to run same. In other words, what HAS to run vs. what can you live without? For example, for me there is no point trying to run high-speed TPC/IP communications with the outside world during widespread power outages as Comcast automatically shuts them down.
For surveillance equipment, the default POE voltage is 48VDC whereas the NUC that does the recording operates at 19VDC. Etc. So sketch out what has to run and figure out what kind of power you need - how many watts at what voltage. That in turn drives system design. For me, it’s 48VDC and 19VDC.
POE for cameras usually runs at 48VDC which I’m supplying from the gateway, a Mikrotik 5009 w/POE out. (I really like Mikrotik for the very wide input voltage ratings for most of its equipment). Meanwhile, my Intel NUC operates at 19VDC input.
Thus, my highest voltage had to be 48VDC and now I need a UPS to run the system. The DCW20 can step up voltages from 12 to about 50 VDC, but ideally, the DCW20 would act more as a pass-through. Thus, I chose 48VDC as a starting voltage.
I also needed power for the NUC running BlueIris as well as a switch that allows the NUC and a few other components to stay in touch. For the NUC and a 12-27VDC switch I added a NUC DCDC power supply downstream from the DCW20 to supply the NUC and the switch with up to 6A at 19VDC.
Other, expendable network hardware is also getting power from the PSU in this build (a Meanwell UHP-200-48) but most of the power it produces goes into a Nextys DCW20, which operates as a small, DIN-rail mounted DC-DC UPS. When I say expendable, it simply means that I do not need it to operate the NVR.
What I love about the DCW20 is it’s simplicity, ease of connections, and compatibility with a wide range of battery chemistries. Allegedly, it’s PowerMaster UPS software will also cause my NUC to shut down when power gets low. That feature remains to be tested.
Despite its Swiss origins, I found the DCW20 surprisingly simple to program. Four buttons do it all, though you can also access more in-depth features via a USB connection. There is also a built-in wizard that you can use. After running for hours while charging the battery, it has remained cool to the touch (it is charging only at 1A, however).
For the battery, I chose a 60Ah LiTime 51.2V battery with communications busses built in. I chose this battery on the basis of multiple reviews and the inclusion of ModBus and RS485, thinking it could communicate with the DCW20. This was likely a mistake, the RS485 / Modbus port on the battery is only apparently intended to connect to a display, while the DCW Mobus seems destined to talk to computers, not batteries. (The Modbus / RS485 connectors shipped with the LiTime battery appear like shrunken, and very proprietary versions of yesteryear DIN stereo connectors.)
Anyhow, as with any power buses at lower voltages, some attention has to be paid to the ampacity of the wires you’re using. Ideally, use fat pure-copper wiring, especially if moving lots of power at low voltages. Also, do not neglect to fuse your connections, no matter how fancy the BMS running the battery is supposed to be. 12AWG is the maximum the connectors for the DCW20 can comfortably manage. I like connectors with heat shrink built-in.
Then there is the issue of distributing the power, i.e. using power distribution blocks and common returns for DC power. I’ll hopefully get into that tomorrow when I’ll try to post some pics. Bottom line, it’s simpler than you think and it saves some money since the UPS market has yet to embrace providing UPS’ with multiple DC output voltages.
Here is a sketch up front of the system. It’s mounted on a 12x12” plate that was recommended by meanwell for the PSU. I have it offset from the wall it’s mounted on to allow convection on the front and rear of the plate. It’s barely warm.
The PSU is mounted to the left, the DCW20 is at the center top, the NUC PSU is below, and the 48VDC power blocks are to the right.
This is a work in progress pic with the NUC and the switch not yet connected.
I chose the UHP200-48 because it’s been my observation that most PSUs only achieve their nameplate efficiency at 50-75% of capacity. So it’ll take about a week to recharge the battery. Which is OK for a UPS.
I purchased a 3x4 acrylic piece of plastic to put in front of the NUC power supply to minimize possibility of shorts.
The NUC power supply has built in standoffs that are M3, so easy to mount neatly if you measure / mark / drill carefully. If you’re not careful, an old milling bit can make the necessary allowance.
Yes
How many watts is that equipment going to pull? 100w total? If that, whilst working at a baseline rate?
A 60Ah 51.2v battery (16 cell LiFePO4?) will run that for days no trouble. Have you looked at Pylontech US2000/3000/5000 batteries? I understand the information you want can get pulled from them directly via modbus, or even building your own battery (the individual cells are relative cheap tariff malarkey notwithstanding) with an off the shelf BMS?
It’s around 100W total i reckon but not all of that is connected to the UPS. To maximize runtime, you really only want things “on” that are needed to get the minimum job done. So one of the switches attached to the Meanwell will shut off if the power is cut.
Given that my CPU + Network assets consume about 1/4 of the total house power, I want to improve energy efficiency further.
For example, I’m exploring getting rid of a USW-Pro-8-POE 8 port POE switch I have since it seems to be radiating about 20 watts of heat even though it mostly loafs along with minimal load. I have found a replacement switch that has an external PSU and so I hope I can make it run off the Meanwell PSU as well.
Similarly, I’d like to replace my cloud key 2 since it also is way too warm to the touch. Despite the significant work that will be required, I’m planning to move it to a iXsystems app. It would be so great if the docker app could simply adopt a physical cloud key 2 backup file. But that is not to be…
I’m not super impressed with the powermaster software. If I am reading the manual correctly, it does not allow shutdown demands on the basis of state of charge. Instead, shutdown sequences are initiated once a timer has run out - up to 3600 seconds, i.e. 1 hour. Not all that useful for a multi-day UPS application.
Other aspects are going well though I have to overcome some minor wiring issues. System efficiency has increased about 20 Watts ever since I pulled the Ubiquiti switch and replaced it with something smaller even as the run capacity of the UPS went from ~1 hour to days. The NUC is also running off its new PSU as well. That leaves a switch to be connected.
Switch is now also running off the NUC power circuit (19VDC).
The PowerMaster software is too dumb to use state of charge (SOC) as a threshold but thankfully, the DCW does have two dry contacts that can be used to signal a low SOC. So there are options beyond trying to monitor the DCW20 via modbus, which looked more challenging to implement.
Thankfully, the DCDC-NUC power supply features a Ignition voltage sensing circuit which can be used to start and stop the NUC via USB. If the DCDC-NUC PSU is set to automotive mode, a delay can be set from when a signal to stop or start the NUC.
So, program the relay on the DCW20 to be closed if the battery voltage is above the critical voltage. Then cut the ignition wire on the wiring harness for the DCDC-NUC power supply and extend it to the dry relay contacts of the DCW20.
Whenever all is well, the NUC will boot and stay on. When the battery levels drop to critical levels, the relay opens, the ignition voltage goes to zero, and the NUC shuts itself down via USB. Presto, the thing will shut down based on SOC, not just a timer.
I feel like the technology in typical commercial UPSes has barely improved since the 1990s.
Thats because it hasn’t really
There’s probably a market opportunity there…
Lack of standardization re: DC input voltages doesn’t help either, ditto stances by some OEMs to only offer AC input into their equipment. For a DCDC UPS approach to work, you have to make conscious equipment choices. MikroTik is likely the exception that proves the rule, ie see the RB5009 that can run off 3 sources of power.
Then there is the relentless tiering of UPS features. There is no good excuse for NUT not to be a standard feature.
Never mind the obsession with razor blade economics re: getting replacement batteries as they wear out. Standardizing around AGM motorcycle batteries helped the industry drive cost down and minimize maintenance while giving end users complete flexibility re: battery orientation while in use.
LiFePO batteries are the next logical step, lighter, more energy dense but also far more complicated in some ways (ie built in BMS, etc). Unlike AGMs your battery voltage ideally reflect the desired output as series setups common in lead acid lines is discouraged for LiFePo unless one BMS can measure all the cells at once.
Ditto DCDC operation as staying low voltage DC avoids the obvious conversion pitfalls, no matter how well the DC-AC-DC conversion steps are implemented. The underlying technology is also a bit more recent, ie wide input DCDC switch chips.
Anyhow, there are obvious early adopter kinks to deal with, as the simplicity of a APC power chute software package is replaced with a bunch of wires, a USB connection to the micro USB port on the DCDC NUC PSU, etc.
I don’t even mind the lack of standard DC inputs; in almost all cases AC is what’s going to be available for significant power use. Granted that does introduce losses in both conversions, and that’s getting a little off of the purpose of your project here.
Agreed. An ESP is something like US$2 each in consumer quantities (on a dev board no less); in bulk they’re far less. That’s plenty to handle monitoring and reporting of the status of the device (and, heck, to do it over WiFi if desired).
More energy-dense, yes, but the big wins of LiFePO4 (compared to lead-acid) are:
And unlike some other lithium chemistries, they don’t have a habit of exploding.
One of the more frustrating points to me is that UPS manufacturers like to spec max power output (in VA, because it gives a bigger, albeit almost meaningless, number), and they work very hard to bury the watt-hour spec. I want to power my gear for some time, not for “just long enough for it to power down cleanly.”
That is precisely the issue for me, the bundling / scaling of output power with capacity. Find me a UPS with a 200W output rating that can run a day. They don’t exist as best as I can tell, you have to build your own.
Oversizing re: capacity then leads to excessive losses in conversion, just like our server PSUs. Observe the proliferation of effiency tiers in PSUs, especially at the low end of output capacity. Ie the main difference between a platinum and a titanium PSU is at 20% and below output.
FWIW, I did design the network around here with a eventual conversion to DC in mind. It’s one reason I like Mikrotik and POE, it makes it relatively easy. Also, home built UPS’ don’t have super annoying buzzers built into them or at least give you the option of turning it off (DCW20).
A side bonus is helping tame the “cable salad” as the Germans would say. I will keep the labeled wall warts for each piece of converted equipment in case the Meanwell PSU dies but there are far fewer cables floating around now. Especially since cutting them to length is easy now vs. dealing with it in your network stack.
The lack of standardization is pretty annoying in this industry. I wish they could all agree to one standard, make it easier for everyone to connect to it, be it via WiFi, USB, or RS485. Instead, a bunch of folk keep insisting that setting up walled gardens will somehow be the path to UPS supremacy. It really makes you wonder if there is a opportunity to disrupt that market by offering modular components to easily provide multiple bus voltages, etc.
I ditched my UPS after installing a Tesla Powerwall in my home. My rack is part of the critical circuits when the Powerwall kicks in during a power outage
