On-Line Double Conversion UPS project

[THORs_Hand]

Hello everyone!
Today I started a rather important and big project (Having in mind that I don't have a lot of experience in the field of electronics design)
I want to design, build and use an UPS with double conversion, using all I kno about designing electronics, safety in electronics etc.
I do have my rough specs, but I write this post in order to ask for advices along the way (any update on this project will be in this post)
rough specs are:
Double conversion, with integrated bypass witch can be triggered manually or automatically either for maintenance or because of technical problems
2kVA power output in all conditions (battery or rectifier powered) (minimum)
>95% efficiency on rectifier
>80% efficiency on battery
ATX board form factor
temperature, voltage, current, power factor, apparent power monitoring
front lcd with tons of info, USB control and data logging
(Not yet sure) ethernet enabled with network interface displaying all the info there is.
I'm not willing to cut any cost on the build, safety is no. 1 priority, I want to trust that thing with my life.
In the end it will power 3 servers and some random network infrastructure
I could buy one, sure, but why not making it myself?
Well first, I got plenty of time, and the thing that I love doing the most is routing boards, although it's the easiest step, I also have some specs that I want to follow so it's not just my engineer ego 
I could also buy modules, but where's the fun?
Why 80% minimum battery efficiency? because I plan to have 24/36V batteries, I want to step voltage up, and then use a push-pull H bridge (or half bridge) MOSFET (or IGBT's) to create the sine wave needed for output, ofc, I live in EU and our standard is 230V@50Hz the rectifier part will be 230VDC but I don't want to have 3 tons of car batteries to have 230VDC by default, besides that, I need a stable 230VDC for the inverter, after all if my batteries will hold for 2h I want 230V on the output all the time, I don't want to rely on the dropping V to cutoff the inverter, also I expect (obviously) quite a bit of current to be drawn from the batteries, ~83A in an ideal case, but I am quite sure that the current can go up to 85-90A easly, but the batteries I found (from rombat if anyone hard of this company) can provide this current easly.
Currently I am thinking about a control method for the inverter, wich is the most important part, I could go all jelly bean and analog or I could go for specialized IC's I would really apreciate your advice on this one.
I had the pleasure to see all the guts of 2 industrial UPS units, one 20kVA AROS unit and one 200kVA SOCOMEC Delphys MP Elite+ unit (my whole job is dependent on those delphys units, i work in a datacenter) The AROS unit uses all jelly bean logic and the heart of it is an ATMEGA 2560 (the same one used on Arduino mega)
the rest being discreet 74HC logic and opamps + quite a beefy analog devices A/D converter, the SOCOMEC unit uses a ton of digital stuff, barely any jelly beans, the magic is done by a XYLINX FPGA, both were very very impressive in all aspects, each in it's own way i'm not aiming for that kind of complexity but I do aim for that kind of build quality.
Thank you all for reading this, I really appreciate your advices and

[eliocor]

I warmly suggest you to read the IEC/EN 62040 norms;
if you want to interface your device, take a look also to IEC 60950

[NiHaoMike]

The loads are DC, so it makes sense to make it a DC UPS. Look up "DC data center" for more information. It is very helpful for reliability since the inverter is quite complex and a frequent source of breakdowns.

[jbb]

The loads are DC, so it makes sense to make it a DC UPS. Look up "DC data center" for more information. It is very helpful for reliability since the inverter is quite complex and a frequent source of breakdowns.

I worked on one of those once.  We were using some very big load resistors to test it.

I agree that the DC-AC inverter is a load of extra parts, and supplying DC to the load is tempting.  But there are 2 things to look out for.

• Switching and protection devices (mechanical switches, circuit breakers, fuses) may not work with DC. They have a tendency to form unquenchable arcs and catch fire.
• The Power Factor Correction (PFC) circuits in some power supplies may become angry with DC input. (Not clear if this is a real thing - I've only ever heard people being concerned about it.)

(For some servers you can buy DC-input power supplies, which is nice.)

2kVA power output in all conditions (battery or rectifier powered) (minimum)
>95% efficiency on rectifier
>80% efficiency on battery

95% efficiency on rectifier isn't too bad. You can do better if you like.
What about the inverter? 95% there?
For the battery feed you should shoot for at least 90% (probably 95% again!) to save yourself a ton of heat.

Have you designed and built SMPS before?  This is quite a challenge.

[David Hess]

Here are a couple of things I have noticed in modern online UPS design:

1. The input boost power factor correction stage produces high voltage DC which directly feeds the inverter through bulk capacitance sufficient to limit power line ripple.  If this stage is hardened, then the UPS may serve to operate as a line conditioner without the batteries.
2. During operation on battery, a high power non-isolated boost converter or maybe transformer inverter steps the battery voltage up to the high voltage DC used by the AC output inverter.  A transformer inverter might be better but I am not sure how current limiting could be implemented.
3. So the online power path does *not* run through the relatively low battery voltage and the battery charger is relatively low power because it never supports the AC output.
4. Battery voltages start at 48 volts (4 x 12V) or sometimes 60 volts (5 x 12V) even below 1000 watts.  24 volts just makes things more difficult because of inconveniently sized wires and if you need batteries in parallel to meet the current requirements, they might as well be in series.

[NiHaoMike]

4. Battery voltages start at 48 volts (4 x 12V) or sometimes 60 volts (5 x 12V) even below 1000 watts.  24 volts just makes things more difficult because of inconveniently sized wires and if you need batteries in parallel to meet the current requirements, they might as well be in series.

I have never seen that on any common "small" UPS. Is that for specialty units designed for long runtimes?
I have seen a UPS that took a 72V battery pack, but that was a 3kW rackmount unit.

[jbb]

I would recommend 48V battery bank because it will should within Safety Extra Low Voltage limits and will need half as much copper wire area as 24V

To go from 40V (nearly discharged + wire and fuse drops)  to say 380V DC is almost 10x voltage gain, which indicates that a transformer is a good idea. Also this will isolate the battery terminals from the mains which is in my book a very good plan.

It’s important to keep the 100 Hz ripple energy out of the batteries, so a large 380V DC rail cap will be required. This will help absorb spikes and dips on the line side.

Additionally, I expect a 2kVA UPS would have 10A nominal output but need a surge capacity up to say 60A (check that IEC standard!) to make sure that downstream circuit breakers work. The inverter should be able to keep operating after some kind of downstream fault too.

[David Hess]

4. Battery voltages start at 48 volts (4 x 12V) or sometimes 60 volts (5 x 12V) even below 1000 watts.  24 volts just makes things more difficult because of inconveniently sized wires and if you need batteries in parallel to meet the current requirements, they might as well be in series.

I have never seen that on any common "small" UPS. Is that for specialty units designed for long runtimes?
I have seen a UPS that took a 72V battery pack, but that was a 3kW rackmount unit.

All of my small Powerware or Eaton online UPSes are 48 volts or in the case of my smallest Powerware units, an inconvenient 60 volts.  They all support external battery cabinets though.

I *have* seen some newer 1000 watt range online UPSes which use a pair of 12 volt 9 amp-hour batteries but I suspect they are current limited on the batteries.

Lower voltages are also inconvenient on the circuit side at these power levels.  At 2000 watts, I would not even consider 24 volts until I had it working at 48 volts.

[THORs_Hand]

Update nr. 1 on the design:
Instead of using simple mains rectified DC, i'm going to use server PSU's.
Modify the case so it can accept 4 server PSU's.
Why 4?
One powerful one is quite expensive and doesn't really meets my requirements, having 4 (2 series 2 parallel) offers me redundancy, the big powerful one: if it fails it fails and when the battery is gone, good bye 100% uptime, if i have that arrangement of 4 psu's i have redundancy, witch means that if one of them fails the remaining 3 can handle the load, they're robust little things. I'll also implement  load switching capability, when one fails it turns off the devices that aren't connected to a "fail-safe" socket, and the ones in that fail-safe socket won't know what happened. although depending of what PSU's i end up using i might not need to implement load switching. In the case of a power fail and not a PSU fail, nothing will be turned off (otherwise what's the point of having an UPS in the first place).
Now, i am a man that really enjoys precision, if something measures something i want to be able to trust that reading, so, i will also implement some sort of hybrid-multimeter into the thing, with high update rate, and precision components (hybrid resistor dividers, GDT's etc), i haven't decided yet the chips i'm going to use, there are a ton of them. Also, i moved on from a single AtMega2360 main controller (for display, web, serial and measurements like temp) and i move to the same AtMega2560 in combination with the famous STM32. The mega will gather all the data (temperature, voltages, currents) from different sensors, via an I2C bus (so i can be able to control how the things are read) and send it over to the  STM32, this one will handle now the Ethernet, Serial, Display, Front pannel controls, and different triggers around the system, just because it's faster, and if it's faster i have less chances of it going thermonuclear. this might not be the best solution, but i can always change it later

[schmitt trigger]

Today I started a rather important and big project (Having in mind that I don't have a lot of experience in the field of electronics design)

My humble opinion:
If you are an individual who, by its own admission, does not have experience in electronics hardware, you don't start by designing a 2 Kw double conversion UPS. Which by the way, will power a high availability and critical load.

You could start with either of two options:
- A lower power and far less critical load. Let's say about 400VA to power a normal PC.
- if you want to proceed with this particular server UPS, then develop the monitoring, diagnostics and web connectivity portion of it. That in itself is a worthwhile and challenging project.

A malfunctioning supply could cause hardware damage.
I ignore the purpose of the server, but if it is used for payroll, accounts payable/ accounts receivable, sales, or any other transactions involving money, I would not be brave enough to implement a home-brew UPS.

[IanJ]

Today I started a rather important and big project (Having in mind that I don't have a lot of experience in the field of electronics design)

My humble opinion:
If you are an individual who, by its own admission, does not have experience in electronics hardware, you don't start by designing a 2 Kw double conversion UPS. Which by the way, will power a high availability and critical load.

You could start with either of two options:
- A lower power and far less critical load. Let's say about 400VA to power a normal PC.
- if you want to proceed with this particular server UPS, then develop the monitoring, diagnostics and web connectivity portion of it. That in itself is a worthwhile and challenging project.

A malfunctioning supply could cause hardware damage.
I ignore the purpose of the server, but if it is used for payroll, accounts payable/ accounts receivable, sales, or any other transactions involving money, I would not be brave enough to implement a home-brew UPS.

I agree.  Start with a much smaller UPS project and you will learn a lot not only through designing and building it but also how to test it and how to measure it's limits and what happens when you exceed them........and all this at much lower power levels.

Be safe.

Ian.

[THORs_Hand]

Actually, it's not like that.

The UPS won't be home brew it's just in-house design, my job pays me to design it because of some specifications that you cand find them but the're expensive i do have the resources needed to finish this project.
I know that some of you might argue that for this kind of project you need a serious level of system integration (controller-wise) but, i take the simple and the most serviceable way.
I want it to have a relative medium level of integration, so that i case of a fault it's easy to fix.
The server PSU wasn't quite my ideea, the project manager proposed something like this and i was ok with it, we can integrate it i do like the idea though.

[amspire]

I did work at a UPS design company for a number of years and we did make a really good quality online UPS.

First thing - it is definitely not easy. There are better parts available now, but the reality is much harder then the theory. In theory, a UPS is very simple.

First, the ferrite transformers and inductor designs needs a lot of skill. You could easily have to go through 20+ prototypes. You want to use the smallest core as possible to minimize the magnetic path length and the wire length and thickness. If you are forced to go to the bigger core, you need thicker wire and before you know it, you have gone up several core sizes. There is the trade-off between core gap and number of turns needed. There is all the clearance details needed to meet the safety specifications.

Next you have to get the circuits reliable and yet they are very difficult circuits to test. You can get voltage slew rates of 10 billion volts/second. It makes it very hard to see subtle problems in the gate drive. You will probably want something like a Tektronix current probe, great quality differential high voltage probes. Make sure you have a very respectable budget for test equipment - do not be a hero and say you can do it all on a $400 scope and another $400 of other test gear. If you can have a $10,000+ equipment budget, you are much more likely to do a good job and to get results in a faster time.

You may be lucky, but do not be surprised if you have to go through 100+ blown mosfets. Make sure that during testing, you are protected from mosfet and diode shrapnel.

Be totally rigorous with static handling procedures from receiving, through construction and into testing. Badly handled mosfets can blow up, but often not immediately. Often after a few days, a few weeks or a few months.

Then you have to get into the detail. For example, you need to be able to switch between the UPS output and the mains, so you need a phase locked loop to synch the output to the mains. However, if the mains gets too far off frequency, you probably want the output to drift back to the proper frequency until the mains gets back to normal. This can be done on a micro in software, but it takes time to get right. Getting the output PWM loop stable with all types of loads is definitely a lot of work. You do have to build high powered loads of different sorts to fully test the UPS.

If you do succeed, then you have to get it tested and that is not cheap. There is the lightening pulse tests on the input, RF radiation, etc.

Separate ICs for UPS management and monitoring/comms is an excellent idea. The management IC if it fails can easily blow up the UPS - it has to be as close to 100% bug free code as possible. The monitor/comms IC can have bugs without much consequence- you can gradually improve the code until everything works fine. Our UPSs had 3 micros back in the late 80's as we dedicated one to sinewave generation and phase locked loop. Microcontrollers have improved since then. A FPGA is not essential, but it can reduce the component count. You just have to be very aware of the state of the outputs as it boots up into as you do not want all the Mosfets turning on as the FPGA or micros boots.

A good reliable UPS is a very big project. For one person to do everything, you could be looking at a few years work. If you try and claim you can do it all in a year by yourself, you will be putting yourself under a huge amount of pressure and you will almost certainly fail by a long margin. It is a multi-person project.

[schmitt trigger]

Ok, so I won't attempt to dissuade you then.

So my advice, in addition to amspire's advice on test equipment, is that you get yourself a thermal imaging camera.

On a tight enclosure, heat management is an art and science, sprinkled with a little black magic.
Unless you have access to powerful finite-element-analysis software, your best bet to view heat flow is a FLIR camera.

Hot spots can appear anywhere, some you will able to predict but many others you will not. Definitively not.

If you get yourself a FLIR camera, I recommend a simple experiment: check how oatmeal cereal cooks in a microwave oven. Even with a rotating turntable, hot spots appear in the most unexpected places.

[amspire]

Ok, so I won't attempt to dissuade you then.

So my advice, in addition to amspire's advice on test equipment, is that you get yourself a thermal imaging camera.

We didn't have thermal imaging cameras back then, but we absolutely would have bought a couple today. Checking temperatures was almost a daily obsession in debugging.

We also had to model heat conduction in the aluminium heatsinks to check we had the best thermal design. It is worth the effort and you may have to get a custom-designed extrusion. The camera would be great for testing and developing the heatsink design. It is easy to have a heatsink where the fins are actually too thin and so half the fin length is hardly doing anything. Or the heatsink base is not thick enough to take the heat away from the mosfets.

You can take the easy approach and just make everything oversized, but it is very easy to end up with a large, expensive and very heavy box.

[schmitt trigger]

Another black magic aspect is airflow.

I remember that we could lower a heatsink peak temperature about 5C just by moving the fan a couple of centimeters away.

Granted.........5C may not sound as much, but when you are close to peak die temperature it is the difference between a reliable and an unreliable product.

[amspire]

Another black magic aspect is airflow.

I remember that we could lower a heatsink peak temperature about 5C just by moving the fan a couple of centimeters away.

Granted.........5C may not sound as much, but when you are close to peak die temperature it is the difference between a reliable and an unreliable product.

Yes, there is a big difference in heat conduction between turbulent and laminar air flow. If you can get turbulent air flow, you get much better cooling.

And if you just spread the mosfets evenly down the heatsink, the first mosfet has 20 degC cooling air. The last one has 50 degC cooling air. It is nice to think the heatsink spreads the heat evenly but it doesn't.

[ovnr]

Well, it sounds like you're in over your head. Post pictures of the flaming remains, though.

One point: 230V AC is 325Vpp. Given the losses and the fact that 0% and 100% duty cycle tends to get hairy, you should shoot for a 350-370V DC bus.

Also, you mention wanting to use a pile of large server PSUs, and also 95% efficiency when not on battery: That means either running in bypass mode - which means you're not doing double conversion. If you're dropping the mains down to (shudder) 24V or 48V, you're looking at 90% efficiency there, and then you need to boost that up to 350V DC - just like from battery mode. And if your boost circuit is only 80% efficient, you're now at 72% efficiency from the wall. Add the losses in the H-bridge output driver and filter, and I'd expect you're down to 65-70% efficiency.

And a 2 kVA output UPS at 70% efficiency is dumping 850 watts of heat out all day long if it's at full load.

On voltage: 48V is the absolute minimum. I had a 5 kVA UPS that ran 240V DC (20x 12V 7Ah batteries), with a big-ass boost converter.

[NiHaoMike]

I'm going to guess that he works in the engineering department of a data center and the 3 servers are part of a pilot setup they use to validate new ideas. Calculate the cost of building a reliable inverter and you'll quickly figure out why DC just makes sense.

[jbb]

Yeah, DC input supplies do save a couple of steps.

DC battery => Inverter => AC => PFC => DC

Does seem a bit thick...

However, DC distribution does require properly rated equipment (power supplies, fuses, circuit breakers, switches, meters etc.) which can be hard to find / expensive.

I expect it will take over data centres eventually, but the home / office stuff will remain AC.

[Gregg]

DC makes sense on many levels:
It is as reliable of a backup solution as you will find; one of the main reasons the telcos have used DC from their beginnings. It is always on line with less to go wrong. 
There is no switching required when the charger power is lost; the load seamlessly transfers to the batteries instead of being supplied by the battery chargers (called “rectifiers” in the telco world).  Most efficient UPS systems pass almost all of the power around the inverter but the inverter runs to keep synch with the line until the AC line drops and then you can get very unpleasant surprises if the inverter can’t handle the sudden power draw.
Flooded lead acid batteries are still hard to beat in this application.
A good DC system is very scalable. Batteries can be paralleled for redundancy and more chargers can be easily added for more capacity.  Small hot swappable SMP charger modules with a common controller to parallel and share the load are quite efficient as they do not have an added inverter stage to take away efficiency.  Redundancy is far easier to attain with a DC system as more load is added; redundancy is a big key to reliability.
Even UPS systems need batteries; they are still a necessary component. If the equipment is capable of using DC and you want the simplest, most reliable and scalable backup, DC should be at the top of the list for consideration.
If you want to show your inventiveness and talents by designing and building a double conversion UPS, you’ll have a hard time beating the commercially available systems.

[NiHaoMike]

For smaller scale, low voltage DC like 12V will work nicely. Most home network equipment even runs on 12V.

[Circlotron]

Have you designed and built SMPS before?  This is quite a challenge.

I worked as a technical assistant in a SMPS design lab for 10 years about 12 years ago. It is an area that really separates the men from the boys.

Speaking of men, your approach has to be rigorous to have any chance of succeeding, and seeing according to one "expert" rigour is a negative white male trait you may as well not even try. Damned if you do and damned if you don't.

[oldway]

Excuse me to say it but your challenge seems to the limit of ridicule.
It is obvious that you have no experience in power electronics to defend such an idea.

Start by making an UPS by copying a comercial model, with schematics and all the service documentation at your disposal.
Make your own transformers and inductors.

If you succeed to build an UPS that is working properly and as reliably as the comercial model, you might want to consider using the power section and adding your own control circuitry.