DIY Franken-Inverter

[bwinzey]

I've been toying with the idea of making a -very- high-powered inverter for some time now. What I've had in mind is basically a very DUMB system, using the most basic, off-the-shelf parts possible, even used ones that have gone obsolete.
The idea of it is very similar to that of huge, hundred's of kW's UPS's, except much 'dumber', allowing for adaptation

My idea is to have something so basic that it just works. IGBT blew up? no problem, I have 4 more spares, maybe replace the gate drivers and what the hell an EGS002 board is worth what, $3? Less if you buy a couple? And we're back in business. One of those fancy inverters blows up on you and you might as well buy a new one cause you'll never figure out how to fix it. This? It's basically Mad-Max ready. As long as the sun is still shining we'll have power, baby. Just make sure to keep the spares wrapped in tin foil and grounded!

By that I mean, we have an H-Bridge driven by your off-the-shelf EGS002 board, using second hand IGBT's (most are from UPS's as well) allowing for massive currents. I recently purchased 6 TOSHIBA MG300J2YS40 Half-Bridges that can handle 300A each at 600V. These cost me $15 a piece, which is actually dumb cheap considering their specs. (and how much a similar part would cost new)

Sure, they are aged, but given the amount of design that UPS's have, I doubt they were even stressed much more than even 20% their rated specs. I understand some people's concerns with buying used parts, but in MY opinion these are a good buy.

(Source: I work with MRI's+X-rays which in many cases are connected to 250kW+ UPS's. Most of the time when they buy a new MRI they also buy a new UPS as a bundle package, the old UPS then is basically worth scrap as it's extremely expensive to uninstall, transport, then install, and most of the time the batteries are in bad condition (they never cycle them, just sit at float voltage for 10+ years), just in batteries they'd spend over $100k, so they'd rather send them to scrapyards or give them away for free. Then these ebay sellers take them apart and sell the parts. I have yet to hear of a UPS stop working or be damaged in any way besides batteries being dead.)

I don't have much experience with inverters, but I do have experience with Arduino's as well with all sorts of random circuits, both tiny and huge, and of course high-voltage.

But anyways, the basic idea is: we have a ~350VDC reservoir, this could be either some capacitors or a battery, or a combination of both. Off-the-shelf EGS002 board sends the gate signals to some higher-current totem pole drivers, which switch the gates of the IGBT's. 2 of these half bridge modules make a full 300A H-bridge. With de-rating a pair of these should EASILY be able to handle 10kW continuous and 20kW peaks all day, and even then you could parallel them and have a literally bulletproof output stage. (I know some people also dislike paralleling IGBT's but that's how they do it in the UPS's, and I trust them)

The reservoir could be a decently sized capacitor bank, (charged by a high-frequency transformer from a low-voltage 48V or so battery pack) or even better a high-voltage battery (which is what the UPS's do, they have ALL the batteries in series). This would allow you to skip a complete stage of the inverter. This is exactly how the Tesla Powerwall works.

You can then directly charge/maintain the 350vdc reservoir with solar panels very easily with a basic PWM circuit (at 350v, one of these $15 IGBT's can easily handle 100A of solar charging which is dumb overkill, but anyways.

They do sell such inverters https://www.ebay.com/itm/233314195391?ul_noapp=true but that's not my goal personally.


But yeah, sorry for the long winded post. My brain is churning a bit. I am also buying components for a high-frequency transformer like the ones you guys are going for in the DIY inverter thread, to boost 12-48v into 350v. Maybe not 10kW but a decent amount. I got a 100x50x25 ferrite core coming from Ali. It's SUPPOSEDLY Amorphous Nanocrystalline which allows for over 2x the Gauss rating of a normal ferrite core. Also 2.5Lbs of 30awg enameled wire to make some Litz wire. But I have no idea as for what I'm gonna do. I REALLY want to go directly for a ~350VDC battery pack but for now I just want to make the inverter head and a boost converter for proof of concept/emergencies. Also have 30 original HY4008W mosfets for the push-pull, which I don't really want to use all of, maybe 8 in parallel for the push-pull so 16 total?

What really draws me away from doing a low-voltage system is the beefiness and the added complexity of the components required, when one could put all their batteries in series and avoid all of it.

With a decent capacitor bank and enough solar though you could easily pump 10kW through a VERY simple H-bridge with 2 of these IGBT's and another controlling the solar into it. And then the boost circuit to maintain the voltage if there isn't enough solar or at night.

Could easily use one of those huge 40kVA+ 3 Phase 208Y isolation transformers that pop up locally for dirt cheap ($100-200), which will allow you to not only get a split phase system but also add peak power capacity as the huge transformer acts like a flywheel.

You know, the more I think about this the more it feels like I'm just thinking up a UPS except I'll be able to fix it if it ever blows up...



I know this is a long post but it's a lot of different thoughts. If you can help in any way I would very gladly appreciate it. I'm not one of those people that just comes up with an idea and forgets about it, I am invested into doing this, but like I mentioned earlier I have never designed an inverter, so any tips and guidance will go a long way. I'm currently looking to see what I could use as gate drivers for the IGBT's and what other components I may need to have a functioning 350vdc-220vac inverter.

I have begun to search for basic inverter designs using the EGS002 and looking for a suitable one to basically scale up for the IGBT's. Would this be a good starting path?

Current Parts:
6x Toshiba MG300J2YS40 600V 300A Half-Bridges
30x HY4008 80V, 200A Mosfets
5X APT2X61DC60J 600V 60A SiC Dual-Diode modules
1x 100mmx50mmx25mm ferrite core
2.5lbs of 30AWG enameled copper wire
3x EGS002 driver boards
-many little components, resistors, capacitors, diodes, that I can use to prototype.

Need:
-IGBT Gate drivers
-Push-pull driver board. I'm between the EG7500 boards and SG3525 boards. My main concern is overcurrent protection and how they would react to it. Ideally one could have the boost driver react not only to the output voltage but the current as well, that way we could have a wide input voltage like 12-48v and it would be able to adapt to it. Cause I think if I make the transformer a 2+2 primary and 65 secondary it would work perfectly for 12v let's say, but if I connected it to 48v I feel we would need a duty cycle 1/4 that of 12v's in order to avoid drawing too much current. Not sure if any pre-made board is able to do that.

 

[toms]

Good luck with it, keep us posted on your progress ?

 

[bwinzey]

So I was doing quite a bit of research and consulted with the people at openinverter (which is an open source three-phase electric vehicle motor inverter). The gate drivers they used originally were the HCPL-3120, using a DC-DC converter to create both + and - voltages for the gate driver. They turn +5v into -9v and +15v to switch more efficiently. The designer of the inverter responded that these drivers (2.5A rated) Have been used to drive up to 600A IGBT's without issues. The only problem is that the switching frequency of the EGS002 SPWM is ~20kHz, and the openinverter only runs at ~8kHz, meaning that the very tiny DC-DC converters won't be able to handle the 2.5x power flow through them. (They're only rated for 1w).

What interests me about this is that the IR2110's in the EGS002 board have very similar specs to the HCPL-3120, and MAY be able to directly drive the IGBTs, though without a negative gate sink.

The only way to REALLY test this is to set it up raw on a low current ~24v+ supply/in series with a 12v incandescent and hooking it up to the scope. I'll be doing that in the following days and give an update.

Hell, I may be pleasantly surprised. Imagine if the IR2110 could directly run the IGBT's decently? It would be a godsend. I have a feeling it won't since we don't have a negative switching voltage in the EGS002 board, and would just be pulling to ground.

Worst thing that can happen is the IR2110 drivers are overloaded and struggle to switch the gates AND/OR they shoot-through due to lack of pull-down power, it doesn't matter as there isn't enough power to damage the IGBT's. Maybe the IR2110's but those cost literal pennies.

Will update in the next few days.


Here's the gate drive circuit on the openinverter. (source https://openinverter.org/docs/index.html?en_gate-driver,23.html)

 

[Roswell Bob]

The IR2110 will not drive those modules. If you want to test gate drives you will need a double-pulse tester. You want to switch into a recovering diode.

 

[bwinzey]

The IR2110 will not drive those modules. If you want to test gate drives you will need a double-pulse tester. You want to switch into a recovering diode.

So I came to the realization that making a gate drive board would be way too time-consuming for me, especially considering I have very little experience in the field.

So I just decided to purchase a pre-fabricated one, which has a built-in egs002 and controls everything. All that's needed externally is the IGBTs, bus capacitors, and output filtering. It has a TIP41&TIP42 totem pole driver for each gate that can handle 6A continuous and 10A pulsed, which is enough to drive pretty much any IGBT. They mention people have used 1200A ones with success, which I don't doubt as I mentioned earlier that the openinverter project has had people run 600A bricks on 2.5A peak gate drivers.

Cost of the board was $55 plus $11 shipping.
https://a.aliexpress.com/_mPZYPJD

Figured for the money I could apply for a job at McDonald's, get hired, work for a week, and buy a bunch of these boards before I get even close to making a halfway decent one myself.

I still purchased a bunch of components but my main problem with making my own is the isolated power supply that I would need for each gate driver. I have no idea how they do it in the board I purchased, but I presume it has to do with that big transformer on the side. Once it gets here I'll definitely try to reverse engineer it to figure out how they were able to make the board so cheaply. Making it oneself would take $20 In isolated DC-DC converters alone.

I feel like it was the right decision at least for me. I wish I could've come up with some crazy simple way to get the job done but for $66 I don't think it's worth fussing much.

Though, $66+$30 for the IGBTs + $20 for a massive NOS 450V 8400uF capacitor I'm barely $120 in on the almost complete output stage. Plus the filtering but I have no idea how much inductor I will need. And this is an output stage that should handle what, 15kW all day long, with 30kW peaks?

 

[Roswell Bob]

These big modules want the gate drives right on top of them. You may be able to get away with some twisted pair if you have a negative gate drive. keep them short tho. The modules want to be mounted very close together as well. I think you are running the dual modules. You will want a low inductance bus structure. A multi-layer PCB with the plus and minus bus interleaved is best. If not you should consider a DRC snubber on the rails on top of the modules. Did you get buss bars with the modules?

 

[bwinzey]

I agree that the gate drives need to be as close as possible to the IGBT but from what I've seen it's not critical. Most of the examples I've seen have ~4" or so of twisted wire from the drivers to the gates, except those with the IGBT's that have many small round pins to be soldered thru a PCB.

I'll still place them as close as possible but I was looking through pictures of UPS innards and most seem to be OK with some distance. Here's a video I found of some random guy that *I think* installs UPS's, showing the inside of a 20kVA 3-Phase UPS.:

. It's odd that they decided to go with 3x100A modules in parallel per phase rather than 1x300A one. Maybe it has to do with dissipation or god knows how old that UPS is (maybe they didn't have such large modules back then?). But if we ignore 1 phase then we'll get 13.3kVA which is pretty much what I was expecting, considering how much they de-rate UPS's for sake of stability I wouldn't be surprised if it could push 15kW continuously with proper cooling.

I didn't get bus bars but I will eventually make some. By a DRC snubber do you mean those rectangular capacitors that they sometimes put right above the IGBT's, like in the video I just linked? If so, I did buy a pair of 3uF ones, which (I think) are the biggest you can fit on these modules.

And btw, the driver board I purchased does have a negative gate drive.

 

[Roswell Bob]

Looks to be about 40 years old. If that old then before the IGBTs. IGBTs began to take off in the 90s. Probably Darlingtons rather than IGBTs. The drivers and power supplies look like they are good for a few amps of base drive. The input looks like SCRs. So maybe they are charging the batteries with the front end. Lots of circuitry to get the job done. I'm wondering how clean the output looks. They would use dual modules to spread the heat out. A simple capacitor on the bus would cause a lot of ringing. Ringing that would require a bit of filtering on the output. Switching transients are usually snubbed with a DRC - diode resistor cap. Cap charges up through diode when eating up energy trapped in the bus inductance. Then cap discharges back into the bus by resistor and gets ready for next pulse. That is why it is important for a low inductance bus structure. Wires from battery should be short and tightly bound together to reduce inductance. Remember that the inductance is a function of the loop area. Smaller loop area, smaller inductance, less energy that you need to snub. In any case very old design. We do a lot better these days.

 

[bwinzey]

Yeah what surprises me is that that thing's still kicking, ~40 years old, running 24/7 in India of all places, where stable, clean service is nonexistent.

I was looking to see what I could do for bus bars and decided to go the cheapest route, I got 1/2" L copper pipe and put it in the press to flatten it. Ends up being 25mmx2mm (which is exactly what I expected from calculating using the circumference and wall thickness).

According to the charts I've seen online, 25x2mm copper bus bars can handle 250A. Since there's 2 in parallel they are enough for 500A continuous. At 350v this is ~175kW. 10x more than I EVER expect to pull out of it PEAK. As a matter of fact I think I'm only gonna have a 100A breaker on it, like my house, except of course 100A at 350Vdc is quite a bit more kW than at 220Vac. Regardless I could just double up on the bus bars. I got a 10' section for $18, and 4' is enough to make a whole set, so I could use 25x4mm bus bars and still have some left over. Should handle 360A per, so 720A total, which seems more on par with what these IGBT's can handle.




This is too big for the 'small' 300A modules but since I have gone with the pre-fab inverter driver I decided to go big and build the inverter using these humongous FF1000R17IE4P 1700 V, 1000 A bricks that I got a while back. I wanted to first build one with the 300A modules but I figured I might as well build the primary inverter if I'm gonna put the effort in.

And depending on how well it works I'll make another separate inverter with the 300A ones to have as a secondary/backup (even though I have 2 extra of these bricks as well but I imagine that if something blows it's gonna be caused by a failed driver rather than just the bricks.)


Now I'm not sure how I should do the busses. I have 2 ideas pretty much.

or


Not sure how important it is to have the power delivery to be exactly the same to both modules, or if having loops like this is bad or even good.




The bus bars are a little wider than I had hoped, but from all the research I've done they're far enough away that its perfectly fine unless something shorts it. The distance between the bars is at least 1.3CM-1.5CM. I could always try to straighten them out perfectly or if it comes to it, grind them slightly, but from what I understand 1CM should be able to withstand 1KV in the most polluted, dusty-humid air, and 2.5KV in normal conditions, 2x above what this will ever go before blowing up the electrolytics which should blow the breaker/fuse regardless.

I've also purchased a MASSIVE 10"x9" 14lb heatsink with ~81 4.5" fins off of ebay, should be getting here soon. From the calculators I have used, the heatsink should be enough to handle ~500W WITHOUT fans, with the modules never going above 80C. But of course I will add fans. I did the math and at 200A these modules have a Vce of 1.25V, which conveniently means that at 200A they will each output ~250W of heat or 500W total. Which I sincerely doubt these will ever even go that high, not even during a stress test, considering 200A at 350v is 63kW.


All of this is of course EXTREMELY overkill but hey, with the amount of money I'll have into this, I wouldn't even be able to buy a proper 5kw inverter. Well, maybe if I were to stop buying 4x of what I actually need... ?

At this point the whole low voltage section is looking further and further by the minute. BUT I do want to make 2 different modules for this inverter:

1. Backup LV to HV converter. I have decided that I will straight up just make 100S or so battery to directly power the inverter, BUT I would still like the ability to connect to a car for example (the classic HF push-pull ~12-48v to 350v). With the components I have purchased it SHOULD maybe do 5kw but it's not critical. And I think it will be quite complicated but we'll see.

2. Solar PWM controller. MPPT is much too complicated for someone like me to do from scratch. What I SHOULD be able to do with a simple Arduino is run a 600V or so string of solar panels through a single IGBT and PWM it to maintain the 350-400v with proper current sensing. This should allow the battery to be charged (if installed) or allow the inverter to run purely off solar without any batteries, as long as I have either a large enough capacitor bank or enough solar to handle the peaks directly.

 

[Roswell Bob]

I am not familiar with those modules. I did some work for IXYS/ABB when they were releasing their first IGBTs into production in 1992. I then spent a few years at Lenze developing some gear, but left the IGBTs in 2007. Are those Toshiba parts?

Your bus bars are fine. Look at using a stud mounted rectifier on one of the bus bars and a resistor and capacitor for your DRC. Keep the wiring loop very small from your batteries. You should keep inverter close to battery bank, Ty-wrap the cables together to reduce inductance. I assume you will be using a LEM type sensor for overcurrent protection. You want to be able to detect and shutdown in about 5-10 microseconds.

I am doing a 2.5kW/5kW peak push-pull converter module. You can leverage that design, but I would stay far away from 12v input at anything over 1kW.. It is a resonant circuit so that the current is very low when the primary switches commutate. They can be series or parallel for as much power as you want. The LT spice file is up on my website. www.veritypower.com Take a look at it.

MPPT chargers are overrated. If you get your maximium peak panel voltage near your pack voltage then you will do well with a direct connection

 

[bwinzey]

The board finally got here, while I was waiting I cooked up this... Thing.


I tried tinning the busbars, didn't come out too good and I had to sand them flat at the bottom for good contact but anyways, it's done. Also made a temporary LC filter with 2 1MH inductors in series and 8.8uF of caps for testing.

Now, the driver board got here and I slowly went through the datasheet and labeled all the wires properly, quadruple checked everything, and hooked it up to 12v. To my surprise, it DIDNT blow up! And output a beautiful sine wave (after the filter of course). AT 9.2vrms, which sounds about right as I had my power supply set to12.8V.


Now, before testing it any further, there is one thing that concerns me. During this testing, which lasted maybe 5 minutes or so, the gate resistors were getting HOT. I measured it with a temperature gun and got around 80-90C from the four.
Everything else on the board seemed OK. The gate drivers were getting to around 50C, but I'm considering putting heatsinks on them just in case.
Since this was running on 12v, how would it compare if the voltage at the IGBTs was at operating voltage, like 340v? I would very much imagine that the strain on the gates would be pretty much the same, correct? Unless there's some sort of weird leakage I am unaware of

But yeah. The current gate drive resistors are 3.3 ohms. Would changing this value affect how much they heat up? Or should I just try to look for ones that handle more watts? these are 1W I believe.
I know that the gate resistor is related to ringing but I'm not really sure how to measure that. If someone here is knowledgeable enough and knows I could definitely try to measure it myself and post the results.

Some pics of the board for anyone interested in how it works

 

[bwinzey]

So I did a bit of research and decided to test it again and this time scope the gates. After like 8 minutes or so had to shut it down as the gate resistors were getting dangerously hot, like 120C.

From what I understand after a couple google searches, this is caused by the gate resistance being too low?

Here's a couple pics of the scope on the gates. I don't see any ringing unless maybe I'm not looking at it right. To me at least it looks like the gate is turning on too fast?

I would presume that the peak in the middle of the rise is because of too little reserve energy in the gate drive capacitors?

Any input would be greatly appreciated. Right now I'm leaning towards changing the 3.3ohm resistors to either 4.7ohm or 5.1ohm or even more. Looking at the IGBT datasheet the turn on switching losses are around 27% higher when going from 3.3ohm to 4.7ohm, but the graph shows values from 2ohms all the way to 9ohms, so I presume it's safe to do so without much ringing.

 

[Roswell Bob]

Those are good gate drive pics. What is the thing that looks like a big capacitor? Is that a joke?

The plateau in the gate drive is when the collector voltage comes down. There is reverse transfer capacitance between collector and gate that you are charging. It is normal. That flat spot is going to get wider with higher collector voltage. Did you say you were running 12V on collector. This is the point you where ringing will start. The ringing will be evident if you increase your gate drive resistance too high. Investigate that phenomana - should be easy to find. Look at Infinion / International rectifier papers. IR has a Hexfet book that is worth getting hold of if you can. ST micro or IXYS will also be good resource.

Seems the gate drive resistors are running warmer than I would guess. What is the carrier rate you are running and what part is it? Are you running negative gate drive on those IGTs? I will look at data sheet and get back to you on what I think if you answer above.

Your heatsink is basically useless without air. The fin density is too high for natural convection. You will need a couple of ballsy 4" fans on it or a crossflow blower to get it to work to it's capability.

The pic of the trace where timebase is at 5ms confuses me. That is not a sine wave.

 

[bwinzey]

Those are good gate drive pics. What is the thing that looks like a big capacitor? Is that a joke?

The plateau in the gate drive is when the collector voltage comes down. There is reverse transfer capacitance between collector and gate that you are charging. It is normal. That flat spot is going to get wider with higher collector voltage. Did you say you were running 12V on collector. This is the point you where ringing will start. The ringing will be evident if you increase your gate drive resistance too high. Investigate that phenomana - should be easy to find. Look at Infinion / International rectifier papers. IR has a Hexfet book that is worth getting hold of if you can. ST micro or IXYS will also be good resource.

Seems the gate drive resistors are running warmer than I would guess. What is the carrier rate you are running and what part is it? Are you running negative gate drive on those IGTs? I will look at data sheet and get back to you on what I think if you answer above.

Your heatsink is basically useless without air. The fin density is too high for natural convection. You will need a couple of ballsy 4" fans on it or a crossflow blower to get it to work to it's capability.

The pic of the trace where timebase is at 5ms confuses me. That is not a sine wave.

Haha that big capacitor is a 450v 8400uF capacitor. Why? Was it a bad idea or is it just hilariously big?

Did you say you were running 12V on collector.
Yes, but I did test it all the way up to 33v which is my power supply's maximum and it behaved well. I didn't scope the gates on 33v though, I'll do it in a bit and compare the results.

What is the carrier rate you are running and what part is it?
The switching frequency is 23kHz coming out of an EG8010, it seems to go into various optocouplers/isolators and then into a TIP41/TIP42 push pull type deal, which switches +15v and -5v (which can be seen by the vpp in the gate scope being 20Vish). I believe they do this as an economical way of driving large currents, as the TIP41/42 can do 6A continuous and 10A pulsed, which is quite a lot more than most gate drivers out there. From my limited understanding of this system, I think they use the transformer on the board to produce 4 different rails of +15 and -5v from the 12v input.

The entire drive board consumes almost exactly 1A at 12.5v and doesn't change with the collector voltage from 12v-33v. So ignoring any heating in the IGBTs I would assume that around 12.5w are being dissipated throughout the board. From the "guesstimate" given by the openinverter forum, calculated that at this frequency and drive voltage, each gate would consume about 2W. Which I would assume is almost completely lost in the gate resistors? And this leaves 4w or so for everything else which seems fair. But yeah that heating was my concern. Maybe I just need higher wattage resistors. Cause the rest of the board does get warm, like the TIP41/42's were getting around 60C, and the L7824's on the underside of the board around 65C. But not "holy crap this is about to light on fire" hot. with even a little bit of airflow they should cool off quite a bit since they are SMDs after all.

Your heatsink is basically useless without air. The fin density is too high for natural convection. You will need a couple of ballsy 4" fans on it or a crossflow blower to get it to work to it's capability.
I already have a couple 6.5" fans from scrapped ultrasound machines ready to use. Tested with one just out of curiousity and the airflow going out was VERY impressive, especially being only a 12v 0.5A fan. Just eyeballing it I think a single fan should do given how it felt but I'm planning on putting 2 and making some sort of forced air induction enclosure.

The pic of the trace where timebase is at 5ms confuses me. That is not a sine wave.
That is the gate drive of the left IGBT module. It's confusing to me as well how they generate the sine wave, I would've imagined that they would use inverted spwm on both IGBT's symmetrically but this is not the case. The IGBT on the right (marked R) is the one that gets spwm driven. The one on the left, both gates had 60Hz square waves. I'm not sure about the timing as I only have a single probe so I can't probe both gates at once.

I don't quite understand it but evidently it works, I've loaded it to around 4A at like 22v and the output sinewave is still fine.

 

[Roswell Bob]

What is p/n of modules?

The resistors on left side should be cooler if only running at 60Hz. Those gate drive resistors will run warmer as the collector voltage goes up. Are you doing LF or HF inverter?

 

[bwinzey]

what is p/n of modules?

They are Infineon FF1000R17iE4P bricks

 

[Roswell Bob]

ok

 

[Roswell Bob]

i go look up

 

[Roswell Bob]

Looks like about 350ma of current through the gate drive resistor. There is also some current associated with the capacitance from collector to gate and the collector voltage. I got this from LT Spice simulation. The calculations are fairly easy to do, but we can let LT spice do them for now.

I would take a dc power supply and put about 350ma through the resistor and see how hot it gets. Then adjust the supply until it runs the same temp as what you are seeing- you said 120c. If you can keep the resistor in the board so it has the same thermal response that would be good data point.

Run the board up with and without collector voltage and see what the resistor temperature does. It should run a few degrees cooler without collector voltage. Take a look at the step in the gate drive. It should be almost nothing without collector voltage.

If you short the output of this thing your big bus cap is going to dump into your IGBTs. Will IGBTs survive.

Is this a LF or HF design?

 

[Roswell Bob]

Do you have link to open inverter forum?