Post by: ritaismyconscience on April 12, 2020, 04:20:10 AM
I'm planning on building a bigger coil using more paralleled TO-247 parts. Bricks are really expensive for me, I can buy about 150 TO-247 IGBTs for the price of a single CM300 brick and I can buy 85 of them for the same weight (shipping from China is fairly expensive). I have an aluminum heat sink which can fit a full bridge of 7 paralleled IGBTs for 28 in total, and I should be able to switch 1750A peak current. (I checked and the parts have positive temperature coefficient, they'll all be on the same heatsink, and I'll arrange my IGBTs for current sharing). I'm guessing I might need to use a different driver because the gate capacitance is quite big (around 160nF for 7 IGBTs).
Here's what I came up with:
(https://i.imgur.com/Cn9hg46.png)
Are there any problems with my plan?
Post by: davekni on April 12, 2020, 06:01:34 AM
https://highvoltageforum.net/index.php?topic=798.msg5332#msg5332
Start with some Spice simulations of your basic topology, and later with parasitic inductances added. Your biggest challenge may be improving construction techniques, especially parallel copper planes for VBus to the IBGTs and local snubber caps.
Post by: ritaismyconscience on April 12, 2020, 06:45:45 AM
Here's my bus design which should make inductances a bit less than with my other coil.
Post by: davekni on April 12, 2020, 08:11:44 PM
Such boards can be made with knife or dremel-tool cuts in raw copper-clad boards - much cheaper than buying ECBs, and much easier than home etching.
Another suggestion: Consider spending more time using and experimenting with and learning from your first coil before starting another one. Run it at higher repetition rates to see how that affects spark length and tuning. Scope more nodes through your circuit to understand affects of parasitic inductance, etc. My 40-IGBT version was my third coil and sixth drive system. (I'd made multiple iterations of electronics for my first two coils, improving on what didn't work well.)
Post by: ritaismyconscience on April 13, 2020, 05:41:42 AM
I'm probably going to use the same IGBTs that I used last time. I don't know why, but some of them seem to show a voltage drop when tested with my multimeter.
I have enough space on my heatsink for a few diodes too, and I think I can use a few RURG8060 diodes.
Here's the circuit I came up with for driving the gates. Any problems? (Checks out on my circuit sim)
(https://i.imgur.com/BtcTxnr.png)
I'll probably have to determine the gate resistor values experimetnally
Another thing I want to ask about is the ferrite cores. I'm trying to use smaller ferrite cores because they're less expensive to ship. I'm wondering if I can get away with using a ferrite core with dimensions OD 31mm and ID 19mm. The one I'm using right now has OD 48mm and ID 30mm.
Also where did you find IGBTs with 350A peak current rating?
Post by: davekni on April 14, 2020, 06:26:43 AM
For GDT, the requirement is that magnetization current isn't large enough to cause much voltage drop across the driver chip outputs. Meeting this requires reasonable inductance, and especially avoiding core saturation. More turns can compensate for a smaller core, but more turns adds leakage inductance. If you stay with a floating gate-drive buffer, then leakage inductance is less important, as the GDT will be driving much less capacitance.
BTW, IGBT bricks are just multiple parallel devices, IGBTs and diodes, in one package. Bricks typically include individual resistors for each gate internally, and separate source pins for gate-drive return. Driving bricks directly through GDTs is common here, so should be possible with paralleled TO247 devices. However, I have no experience with that, as I have floating buffers. I'm using isolated gate driver chips rather than transformers, however, to feed the buffers. The isolated driver chips add dead-time between turn-off and turn-on.
Concerning your specific ferrite core, I can't say much without specifications.
My IGBTs are rated for 240A peak at 15Vge, including safe hard turn-off from that current. For soft-switching, I'm running them past their peak current specs, and at 18Vge. (Most designs here run bricks a bit past their peak current specs as well.) I'd initially tested one part with 500A pulses. Just finished a more aggressive stress test running one part with ring-down waveforms of 650A peak, repeated at 30Hz. No heatsink, so the IGBT case is 138C. After about 25 hours running this way the IGBT fried shorted. I haven't dissected it yet to see if the IGBT or diode actually fried.
Post by: ritaismyconscience on April 16, 2020, 07:59:09 AM
Here's the interrupter circuit:
(https://i.imgur.com/RVrvhIP.png)
Edit:
http://s000.tinyupload.com/?file_id=01763471399166613607 (http://s000.tinyupload.com/?file_id=01763471399166613607)
^schematic
Post by: ritaismyconscience on April 19, 2020, 10:35:43 PM
I guess a "dumb" iron core transformer would probably be much more reliable.
Also, I don't think a GDT would work. The CM600DY brick has about 180nF gate capacitance. Each of my IGBTs has 30nF capacitance according to datasheet, so 14 of them would have 420nF, which is much higher than the brick.
Post by: davekni on April 20, 2020, 12:14:22 AM
Your IGW75N60T parts are at the high end of TO247 IGBTs for gate charge, 450nC for 0-15 gate swing. The brick you reference (CM600DY) is towards the low end, 2700nC. Others are higher, such as CM600DX at 4500nC.
Brick gate-charge specifications apply to each IGBT, not the pair combined. Compare against 7 of your IGW75N60T parts at 7 * 450 = 3150nC. Using a GDT directly is a reasonable option, as is direct drive as I chose for faster switching. One down-side of at least the simpler direct-drive designs is that IGBT gate drive swings from 0V to +24V (or 0 to +18.5V in my case), rather than -24V to +24V. This uses less gate-drive power, but doesn't add dead-time to prevent momentary simultaneous conduction of high and low-side IGBTs. I use isolated-gate-driver chips, which include adjustable dead-time as a feature. You will need to include dead-time somewhere in your gate-drive circuit.
Post by: ritaismyconscience on April 20, 2020, 12:57:42 AM
Post by: davekni on April 20, 2020, 04:39:15 AM
Rather than four GDTs, I'd suggest a single large one with many twisted-pair windings. It would be ideal if you could use 6 or 8 IGBTs instead of 7. Then your GDT could have 8 outputs driving 3 or 4 IGBTs each, or even 12 or 16 outputs driving 2 IGBTs each. That handles issues with voltage drop on the source leads between the different paralleled IGBTs coupling into gate drive. Of course, it means needing more gate damping resistors and parallel fast-turn-off diodes. (The IGBT bricks typically have multiple internal gate resistors.)
Make each GDT output one wire of a twisted pair. After winding each twisted pair, re-twist the primary leads as a pair to the driver, and the secondary leads as a pair to the IGBT(s). Thus, if you have 16 GDT secondaries, you'll also have 16 GDT primaries, which all wire in parallel to your discrete FET driver buffer. Of course, make sure the phasing is correct for each output.
Although not popular here, EE core sets work well. My SSTC GDT has 8 outputs and 8 paralleled primary windings on an EE55 core, two turns per winding. Here's a couple EBay links for suitably EE55 cores, but you can find other sources too:
https://www.ebay.com/itm/E5521-Ferroxcube-E55-E-EE-EE55-Ferrite-Cores-transformer-AL-6300-1set/251024599965?epid=1372843764&hash=item3a723b6b9d:g:bxsAAOSwF7lb6rmf
and
https://www.ebay.com/itm/Qty-4-FERRITE-E-CORES-H7C1-EE-55-55-2-NEW-2-PAIRS-or-4-PCS-TDK-H7C1-77-23-Z/310352364736?hash=item48427144c0:g:sD4AAOSwEVZaXmTF
Two turns is sufficient on this core as long as your gate-drive buffer has low output impedance (and a sufficiently-large series capacitor if you include such). You could use three turns for margin.
Post by: ritaismyconscience on April 20, 2020, 09:35:59 PM
One problem I see with SMPS is that there's a Y capacitor between mains and the output ground. At 80khz you would pull quite a lot of current through that capacitor.
Post by: ritaismyconscience on May 16, 2020, 11:45:31 PM
(https://i.imgur.com/6GZizN2.jpg)
IGBTs def look suspicious, I tried testing G-E capacitance. You can see the results under each transistor. It could simply be that the one on the left is a newer version of the other two (you can see the middle transistor is real but has a significantly lower capacitance than the right transistor).
(https://i.imgur.com/wkH2oaL.jpg)
Seems to be some sanding marks on the suspicious ones. None on the real ones.
(https://i.imgur.com/eYEJaZO.jpg)
Back side looks different too (3 on top are questionable, two on bottom are real)
Edit:
I found some pretty good evidence of sanding and remarking:
(https://i.imgur.com/NKFUqrg.jpg)
Take a look at the date code markings. Top one definitely is sanded, you can see the ridge isn't as big as the one on the bottom, which is real. Will be messaging the seller. Anyways, I bought these for the same price as the other ones I used in my big TC, but the other seller actually sent legit parts.
Post by: davekni on May 17, 2020, 02:25:05 AM
The low capacitance could indicate counterfeit, or possibly a newer version of a real part. As semiconductor processes advance, it becomes possible to make smaller die with the same current capability. Usually thermal performance suffers, however, especially transient thermal capability. You could search for any updated specification with higher thermal resistance, or dissect one of each and compare die size. Of course, a smaller die could also easily indicate a counterfeit part.
Post by: ritaismyconscience on May 17, 2020, 03:14:25 AM
I also bought some other parts that had the same pattern (horizontal sanding marks) These diodes turned out to be fake as well (capacitance of 204pF vs 760pF for real part, off by a factor of 4). I guess I got lucky the first time I bought, as those parts were genuine.
Is there a simple way to test the current rating? I don't have any fancy test equipment at home.
Edit: sellers insist they are real, usually with fake parts seller just gives up because doesn't want to get a fraud complaint. Seller says that they are sending out newer parts now. Anyways testing the reverse recovery time should be really easy.
Post by: davekni on May 17, 2020, 05:21:44 AM
[ Invalid Attachment ]
Parts can be rearranged too, and values can be adjusted. The basic idea is to make an L/C ring-down test that hits the intended peak current on the first half-cycle. Here I use 20uF and 20uH to make a 1-ohm resonance. Charging C1 to 220V would ideally produce 220A peak, but due to losses may be around 200A. Scope Vce to see if voltage drop is as expected, and check diode voltage on the second half-cycle. Scope voltage on the L1-C1 connection to see the actual ring-down amplitude and frequency. Presuming your capacitor is accurate, frequency will give a measure of impedance. To find peak current, average start and end voltages for the first half-cycle, then divide by impedance.
For the inductor, I usually just wind an air-core coil with some scrap of wire. The capacitor is best to be polypropylene - a motor-run capacitor or DC link or whatever. It doesn't need to be rated for AC use for this short test.
The hardest part is getting a clean switch closure. If you have any TRIACs around, a BTA16 will handle 500A for such a short test, and produces a clean waveform as long as voltage is high relative to a diode-drop. If not, prop/tape/... a pair of wires, one against a board and the other a cm or two above it. Hit the top wire down with a dowel or plastic rod or whatever. The hard fast impact will often make a reasonable fast switch closure. (Using relatively large C and L values for low frequency makes switch closure less critical.)
BTW, I use this ring-down test much more often to test inductors. Same circuit except the IGBT and its gate supply are missing. The ring-down waveform will start with a flat-topped distorted sine wave due to inductor core saturation, then ring down to more of a sine wave. Measure area of a half-cycle, divide by 2, and that's the volt-second capability of the inductor (or transformer winding).
Post by: ritaismyconscience on May 18, 2020, 03:18:40 AM
Seller ended up admitting that they were the 50T60. I tested a real 50T60 part, and I confirmed this (G-E capacitance was 12nF for the 50T60 part, similar to the fake 75T60s).
Ended up agreeing on a partial refund (since I did not want to ship the products back to China).
Post by: ritaismyconscience on May 23, 2020, 04:14:52 AM
(https://i.imgur.com/NuGOtTT.jpg)
(https://i.imgur.com/IZFWzhg.jpg)
(https://i.imgur.com/5vhKhaI.jpg)
Any problems?
Post by: davekni on May 23, 2020, 07:41:41 PM
When I've had to stack capacitors for space constraints, I run copper foil up the sides to reduce inductance of the long leads. You should be fine as is, though, especially since it looks like you plan to bolt a couple electrolytic caps directly to the board.
Post by: ritaismyconscience on May 26, 2020, 02:01:33 AM
Top view:
(https://i.imgur.com/uD0LhxR.jpg)
Bottom view:
(https://i.imgur.com/0IlZzmN.jpg)
A closer picture of part of it:
(https://i.imgur.com/XwvetS8.jpg)
Transistors are actually 50T60 (I wonder if I can run it at 200A each or 1400A total). Also, I'm using 3 x RURG8060 as antiparallel diodes, do I need more?
Post by: davekni on May 26, 2020, 05:24:41 AM
200A is probably OK as long as you have a bit of phase lead and gate drive above 15V (say 18V+). That also presumes reasonable current sharing.
Be careful handling your H-Bridge as it is. Reaching to pick it up and accidentally touching a gate lead before the bulk supply could easily fry an IGBT with ESD.
Post by: Mads Barnkob on May 26, 2020, 08:27:58 AM
Did you do continuity test or high voltage test to check for large enough respect distance, seems there is some copper pieces still along the insulation zone in the middle?
Did you test the board for each IGBT soldered on? Could maybe have used terminals to make exchange of IGBTs easier, I foresee a few blown ones ;)
Post by: Max on May 26, 2020, 03:57:04 PM
Long story: I had to learn the above the hard way. Twice.
I've build two DRSSTC bridges so far. The first one was a 5 layer stack of 2mm Aluminium plates and 2mm polycarbonate plates as insulator and "filler". The stack was hold together with nylon screws which at the same time fixed the 10 bus capacitors and the 4 IGBT bricks to the bus bars. Performance was great, insulation not so much. Since the layers were only 2mm thick and since I didn't take any other precautions the creepage distance was 2mm, too. This was barely sufficient for 230VAC (325Vbus), but it blew up at 400V (566Vbus) - violently.
(https://highvoltageforum.net/proxy.php?request=http%3A%2F%2Fforum.mosfetkiller.de%2Fdownload%2Ffile.php%3Fid%3D14324&hash=b86c5f575b62157979f86f7a06c80e8979101130)
(https://highvoltageforum.net/proxy.php?request=http%3A%2F%2Fforum.mosfetkiller.de%2Fdownload%2Ffile.php%3Fid%3D14325&hash=2708b2d34749c944cc639fe3e0c545b0a7b60047)
(https://highvoltageforum.net/proxy.php?request=http%3A%2F%2Fforum.mosfetkiller.de%2Fdownload%2Ffile.php%3Fid%3D14326&hash=9771f5a4f3519ebc98d37407e1c577ed6369e12d)
So I redesigned the whole thing, this time with more creepage distance. The inner ground layer had bigger holes to get even more creepage to the screw holes. This time I replaced the 2mm PC by silicone. The idea being that with its flexibility it would fill tiny gabs between the layers. All aluminium pieces had 3mm radiuses and all edges were sanded. No sharp points this time. During assembly I cleaned all parts with ethanol and made sure there was not even the tiniest amound of (conductive) dust left over. This, together with the increased 7mm creepage distance "inside" the layers should prevent any problems. Well, I had to reduce it to 4mm on the bottom surface because there was no space for the insulation layer. And guess what: it worked great. Until it blew up ;D . Bottom layer, right where I had to reduce the creepage distance.
Assembly (the easies part actually):
And finally, the most interesting part, when it blew up, and what damage that made:
After disassembling and reassembling everything I managed to add a small bar between those two rails. This seems to work so far, but all in all this bridge design is not optimal. I have new ideas which should be almost as good concerning inductance, but wayy easier to isolate properly.
Kind regards,
Max
Post by: davekni on May 26, 2020, 07:40:48 PM
Also make sure there are no slivers of copper (or other metal) around. I construct many projects with dremel-tool cuts in copper-clad FR4. Sometimes the cuts leave tiny whiskers of copper that need to be cleaned up. Beyond that, at 325V, you are probably fine.
Max,
Are you certain there wasn't some conductive contamination initiating the arc? There are 1200V IGBTs available in TO220 packages with only 1.3mm between leads (and at least one 1600V part).
Line-voltage spacing requirements are high because line-surge capability needs to be ~3kV, and because failure is a potential safety issue. I think 600V is unlikely to jump a clean 2mm insulating gap.
BTW, here's an image of my latest copper-clad with cuts board, after adding labels:
This isn't a high-voltage board, just an example of technique. Cuts are with a small rotory-tool cutoff wheel, with edges cleaned up using a knife and light sanding to finish.
Post by: Max on May 27, 2020, 11:51:19 AM
The argument with the smaller TO packages is good... Didn't think about that. Honestly I can't guarantee you that there was really no contamination. As I said, I wiped everything with ethanol as good as I could.
The best theory was that a transienst initiated the arc over, and the bus capacitors then dumped all their energy in it. The first version had 2*440V = 880V TVS diodes for each IGBT, the newer has four 0.68uF snubber capacitors. After blowing up, too, I soldered 880V TVS diodes to the terminals of each snubber capacitor (only place where they fitted). That setup ran for about 10-20 minutes without problems so far. Won't have the opportunity to test it again til the end of the lockdown.
Kind regards,
Max
Post by: ritaismyconscience on May 31, 2020, 01:03:12 AM
(https://i.imgur.com/r74LXac.png)
Would coating the gap with something work?
Post by: davekni on May 31, 2020, 03:48:54 AM
The bidirectional TVS diodes on the gates is great for making the board survive handling and associated ESD events.
Post by: ritaismyconscience on May 31, 2020, 06:02:30 AM
Also I'm going to use a doubler to power it.
(https://i.imgur.com/s1sKWgr.jpg)
I'm using 8x 2700uF 200V
Post by: davekni on May 31, 2020, 07:23:35 PM
What is your line voltage? I'd thought you were in a 230V part of the world. Peak doubled would be 650V, too high for your IGBTs. If 120V line, the peak doubled is 340V, so should be good.
Post by: ritaismyconscience on May 31, 2020, 08:13:21 PM
Post by: Max on May 31, 2020, 11:59:47 PM
I‘d rather try to omit voltage doubling and stay at ‚easier‘ voltage levels. It should be possible to get similar results by reducing your tank impedance.
Kind regards,
Max
Post by: ritaismyconscience on June 01, 2020, 03:07:36 AM
Post by: ritaismyconscience on June 03, 2020, 12:35:11 AM
Here's my plan: I basically cut out 7 wires of the same length to connect to all of the emitters, and I added 7 1n5819 diodes and 7 10 ohm resistors to each gate. I'm going to connect one output of my GDT to these wires.
Will this work? I'm worried about the emitters being at different voltage levels because of the high current.
Edit: Can I use a quarter watt 10 ohm resistor?
Post by: davekni on June 03, 2020, 07:01:44 AM
If you want to be more sure of Vge, use multiple GDT windings, with each winding driving a smaller set of IGBTs. I'd posted somewhere back in this thread about that option. Even 28 winding pairs, one per IGBT, with all primaries paralleled isn't unreasonable. (The other half of each twisted pair winding are all paralleled for the GDT primary.)
Ideal way to find gate resistor power is with simulation. For a crude estimate, calculate the energy stored in the gate capacitance, then presume that energy is dissipated in the resistor every cycle. (Worst-case, it could be 4 times as high, since the gate is swinging twice the voltage, from -20V to +20V.)
On an unrelated note, attaching TO247 packages to heat sinks with their "mounting" hole is not always effective for heat sinking. The attachment force isn't where the die is located. Most commercial designs use some form of clamp or spring clip applying force over the die, roughly 1/2 way from the leads to hole. Depending on how compliant and thermally conductive the pads are and how consistent screw tightening torque is, this may or may not be a big issue. All it takes is one hot (poorly heat sinked) part to fry shorted, and the failure will cascade. (I think the mounting holes are there for historical reasons. Designs from 50 years ago used the holes, until engineers figured out that thermal performance was unreliable that way.)
Post by: ritaismyconscience on June 03, 2020, 07:57:48 AM
http://www.aosmd.com/res/application_notes/package/AN101_TO220_Guidelines.pdf
On page 4, there is a graph showing thermal resistance vs torque on screw, and it seems like 0.5Nm is sufficient. Also, overtightening the screw seems to make the thermal resistance worse for some reason.
Post by: Mads Barnkob on June 03, 2020, 02:54:04 PM
I searched it up and I found this:
http://www.aosmd.com/res/application_notes/package/AN101_TO220_Guidelines.pdf
On page 4, there is a graph showing thermal resistance vs torque on screw, and it seems like 0.5Nm is sufficient. Also, overtightening the screw seems to make the thermal resistance worse for some reason.
When tightened too hard, the part will start bending from the through-hole and bend upwards around it in a circle, so you start to loose contact again.
Please do remember to have cameras filming when you power up that bridge, its spectacular either way if it runs long sparks or huge explosions :) Multiply cameras from different angles would be great!
Post by: Max on June 03, 2020, 03:04:11 PM
Finally, I'd quadruple check every single signal before feeding any amount of energy in this thing. It looks like it would be an incredible amount of work to repair this thing.
Kind regards,
Max