I drilled and tapped holes for mounting the IGBTs on the heat sink.
I put the IGBTs and diodes on the heat sink.
None of the solid state tesla coils I've seen use a resonant primary. One of the downsides of the SSTC is the large magnetizing current, especially when you try to get a fast voltage rise. This leads to high losses in the power switches and makes it impractical to generate 'normal' rise times and streamers.
Original poster: "jimmy hynes by way of Terry Fritz
My approach is to add a primary capacitor. Pspice simulation shows that the energy transfer is much more efficient. A parallel H bridge of IGBTs drive at resonance (about 70 kHz) by monitoring the zero crossings of instantaneous current with a microcontroller. The current envelope is also tracked to detect the peak of the secondary voltage (minimum of the input current). At this point, the switches are held closed emulate a spark gap before it quenches. This prevents the diodes from sucking power back from the secondary. After one or two beats, the switches are left open as the secondary decays.
I think this approach combines the advantages of a conventional spark gap coil and a SSTC.
Compared to a 19th century conventional coil
Went up to the factory to start winding the secondary. Winding more than two thousand turns takes a long time.
After school I brought the winding project inside to wind and the coil fell of the stands. It damaged some of the insulation and stretched out some of the wires. Dad says take a break, and let someone else to do the grunt work.
Put on a layer of polyurethane to hold the windings in place. I decided to spend my time doing the brainwork and pay someone to do the winding.
I started wiring up the bank of electrolytic capacitors… It’s going to take forever.
Another problem solved: The reason the diode and IGBT cases were shorted to the heat sink is that there were burrs on the heat sink that poked through the insulating pads. The leads of the diodes kept ripping off when I took them off the heat sink.
Notes: IGBT works fine, makes clicking noises at 60 amps. I killed part of a microprocessor by accidentally the gate drive wire to the microprocessor output wire.
Today I tested the IGBT’s peak current handling. I used the microprocessor and gate drivers to pulse the gate at 18 volts for 5 microseconds about once a second. I charged 120uf of motor-run capacitors through a resistor and a diode plugged straight into the wall. I kept the current path as short as possible and used thick wires to keep resistance down.
The capacitor was shorted across the IGBT and a quarter ohm resistor. The quarter ohm resistor was actually two groups of eight in parallel hooked together in series. At about 300amps the voltage drop across the igbt was about 8 volts. At about 500 amps, which is where I want to run it, the igbt did not latch up, but the voltage across the igbt soared up to around 30 volts. That’s an unacceptable loss, I need to try higher gate drive voltage or get some of those electric train IGBTs from ebay J
I tried 3 9v batteries instead of two, and the igbt drivers, microprocessor programmer and microprocessor died. L The driver was only rated for 20v. I will need higher voltage igbt drivers and an opto isolator for the micro for more tests.
Today I tested the break down voltage from the gate to the emitter of the igbt. It survived up to 50 volts (they’re rated for 20). I didn’t test any higher cause 50 volts is probably more than I need, and I might actually kill an IGBT in this set of “destructive” tests.
Today I did more peak current tests, but with higher gate drive voltage. I didn’t get the microprocessor or gate drivers, so for this test I just touched the wire coming from four 9-volt batteries connected in series to the gate. With this test I can’t tell if the IGBTs are latching up because the gate voltage stays high for too long and completely discharges the capacitor.
The test went better than expected, so I shorted the first group of resistors for 1/8th of an ohm load. The voltage-drop across the IGBT was only about 8 volts at 800 amps and at 1200 amps the voltage drop goes to 15 volts. With the higher drive voltage the IGBTs work great, but I have to make sure that they are not latching up. This test went better than I expected and I still haven’t killed any IGBTs!
I decided to use gate drive transformers instead of the bootstrap method. Because these transformers are extremely specialized, I have to make my own. I got the cores from dead fly-back transformers…they were working when I found them. I had to grind out the air gap because I’m using it differently. The primary and secondaries were twisted together using a drill (thanks MOM) to get the k as high as possible. I tested driving the gate with the transformer and it worked but there was some ringing. Now I have to add the other windings on and make another transformer.
I had trouble with the transformer in my high current test. It worked for single pulses only. One of the problems I ran into was very weird: the gate drive was chattering when the opto isolator was not powered. Every time I put the scope probe on the input to the gate driver, the chattering stopped. Kind of like the Heisenberg uncertainty principle.
To get the opto-isolators to fit next to each other on the socket I had to grind some plastic off of each one. Because the capacitor in series with the primary was not charged up, I got twice the normal voltage on the gate. It was off the scope scale, but probably around 60 volts (3x the rating).
The IGBTs did not latch up even at around 1000 amps; I did not test any higher current.
IGBTs are cool.
Today I did another current test. I charged the capacitors up to 8 volts by discharging triggering the IGBT every second or so. For some unknown reason, I left the capacitor in series with the primary out. Once a second was too frequent to avoid saturation. I knew it was saturating because of the clicking noise the xfrmr was making. The gate was still being driven ok, and the driver was ok for a while, so I figured it would be fine. The driver died at the end of my tests. The results are good; the only significant voltage drop other than the resistor is the IGBT.
I’ve been working hard on my Double Resonant Solid-State Tesla Coil. I’m planning on entering this project in the Orange County Science and Engineering fair this spring. Now that it’s official, my parents put this up there like homework!
I’m only a week or two away from making some big sparks. I did decide to go with geek group capacitors and have built my MMC. I have also decided to forget bootstrap charging, and just use gate drive transformers.
I did some IGBT testing. Here is a picture of my test set up--------------------------- The capacitors are three 40uf motor-run capacitors in parallel and the resistors are groups of 1 ohm resistors equaling a quarter ohm, and later and eighth ohm when I shorted the first group. To estimate current, I calculated the amount of current that would go through the resistors with the known voltage across it. I measured the voltage drop across the rest of the circuit to determine whether or not there was any voltage drop across the inductance. The only significant voltage drop was across the IGBT and the resistor. At around 500 amps with 20volts on the gate, the drop from the collector to the emitter was around 30 volts or so. The next step was to try higher gate voltage. I tested the gate to emitter break down voltage. It was fine to 50 volts, and I decided to stop there because I didn’t want these tests to be destructive tests ;-)
On the next test, I just touched the wire from four 9-volt batteries in series to the gate, so I could not tell if the IGBTs were latching up, but I could watch the voltage drop across the IGBT. With 36 volts on the gate, the voltage drop at 1KA was around 7-8 volts. That’s at over six times the rated current (160 amps)! At 1200 amps the voltage drop across the IGBT becomes nonlinear. More recently I did a test using a microprocessor to give the gate driver a single pulse. The IGBT didn’t latch up even at about 1000 amps, which was the highest level I tested. The voltage on the gate was off the scale of the scope but the 40-volt clamp was working, so I don’t think it was much higher than 40 volts. I designed it to push 500 amps per IGBT so I have plenty of headroom. These IGBTs are tough, and work great if you overdrive the gate
Put down some varnish on the secondary, which is now fully wound.
I finished wiring the gate drive transformers yesterday and wrapped them with electrical tape. Today I glued the proto-board and tie wrapped the gate drive transformers to a piece of plexiglass, which was bolted to the heat sink.
Did some preliminary tests once they were on the heat sink: the transformers didn’t work. Almost positive I wired one or more of the primaries backwards which essentially shorted the transformer.
I have a new theory on why the gate transformers didn’t work: On one of them I probably did wind one of the primaries backwards, but even with the primaries in proper phase it won’t work. The primaries didn’t have the exact same inductance and because they are coupled so tightly it is an effective short even with the primaries in phase.
I rewound the transformers. This time I didn’t twist the primaries and secondaries together and I had a single primary.
Did some more testing. Due to flaky connections I had problems with the driver chip oscillating. I saw some unexpected waveforms, but after playing with Pspice, they made sense. I also figured out that the scope isn’t calibrated, it is about 30% high. I have even less voltage-drop across the IGBTs than I thought!
I finally got it working under battery power and with no load. The ringing at the end of the burst is high enough to turn on the IGBTs and I am also getting shoot-through.
The capacitor wasn’t charging because the H-Bridge is shorted. I measured some shorts to the heat sink, so I put two coats of polyurethane on the front of it. It didn’t fix the problem. I took a diode off one of the IGBTs and it was fried. I don’t know what could possibly be killing these diodes because I’m powering it with 150 volts and the diodes are rated for 400. There is also no load, so there should be no current flowing through the diodes.
Tested the remaining half bridge with diodes. This time I used a high impedance resistor between the capacitor and the half bridge. It worked fine with that. I replaced that resistor with 220 ohms and it still worked. I reduced it to 10 ohms and I started to see real narrow voltage spikes. When I removed the resistor I saw voltage spikes up to 400 volts. The shoot- through was allowing huge currents to flow and when the IGBT turned off, the stray inductance caused a voltage spike that was big enough to kill the free wheeling diode. I am going to order 600 volt diodes and more gate drivers so I can add dead time.
To add dead time with gate drive transformers, I have to play some tricks with the timing. If I try to hold the gate down more than up, the transformer will saturate because of the DC bias. To avoid that, I am going to switch the gate drivers like this:
And the gate voltage will look like:
Also, I am going to play tricks with the timing of the voltage pulses on the primary to stop the ringing at the end of burst.
Today the microprocessor is not running fast enough to catch the gate voltage on the way up.
Today I ordered higher voltage diodes, and more IGBT driver chips. I also wired up the socket, so when I get them, I will just have to drop them in.
Today I got the stuff from digikey. I replaced the diodes, which is a lot of work. I also got the other driver chip going, and added 10 turns to the primary of the gate drive transformers, so they could be driven individually instead of in series.
Today I got the microprocessor running fast with the 8mhz crystal, and spitting out the 4 signals I need. I just have to connect them to the driver chips and it should be ready to go.
I tested the diode bridge today. It was one of the few tests I was sure that it would work. I plugged it in, and then… BANG! BANG! BANG! The tops blew off my driver chips, and one of them was shooting red fire out about an inch. It also killed the microprocessor and programmer. Today was a zero progress day; the bridge rectifier works, but I’m a little behind on the driver things now.
I figured out that the other two driver chips are also dead. I did the same test, but with a 20k pot and 50 ohms in series. Nothing blew up, and the voltage output of the microprocessor looked clean as I turned down the resistance. Then I tried to get the gate drivers working in phase again (I had to detach them from the micro for testing, and didn’t label which pins go to which wires). One of the output pins of the microprocessor died, and the gate driver that it was signaling died too. I don’t know why. Now I am out of parts ;-(. I ordered 4 extra gate drivers thinking I was done blowing them up. I’m getting a bunch next time.
Today I got a jigsaw at Home Depot. I cut the circles of Plexiglass for the end caps of the secondary and bolted the two toroids together on a threaded rod using 2” PVC pipe as a spacer. I took a picture of the toroids in the driveway.
I glued the Plexiglass circles together and tried to attach the toroids to it, but the pvc spacers were not cut very close, so the whole thing was tilted.
I replaced the bad drivers, but there was another problem. One of my IGBTs is dead! The wire to the voltage clamp came undone, and the voltage spikes killed it :-(. I was already driving the gates at around 37-38 volts, and I didn’t twist the wires on the gate drive transformers, so the inductive kick could bring the voltage way up. I have had too many problems with bad connections. I also found a short across a diode, the wires were just soldered too close together, and they touched. The wire that brings power to the gated drivers also fell off. Even with no load, and new driver chips, the transformer wasn’t working. I tracked it down, and the primary was open. This time it wasn’t because of one of my bad connections! Although it was caused by one. Here’s my guess at what happened: when the driver chips blew up, the current came through the transformer primary, and the primary wire somewhere in the transformer acted like a fuse. The only point I could find any connection was between the bridge – and the output of the driver chip. I wound another primary, replaced the IGBT, and reconnected all the gate drive wires. I am now back to the point I was at a couple weeks ago, the gate drivers work. I tried powering the diode bridge again. I used a 20k pot in series with 50 ohms there was no waveform distortion through the entire range, with and without power to the gate drivers. I accidentally touched the scopes ground probe to something that wasn’t grounded. I saw sparks. Nothing was hurt ;-). I then hooked up power to one h-bridge with 50 ohms after a 40uf filter capacitor. I forgot to turn the resistance back up to 20k, and I forgot that I was going to leave the gates low for the first test. When I plugged it in, the resistors burst into flames. Nothing was hurt, so it was cool.
I forgot to unclip the ground wire of the scope. it was clipped to the negative part of the bridge, so when it was plugged in it was pulling current half the time. The IGBTs don’t take any current when the gates are low, but I am getting a weird clipped input. I also found out that one of the gates wasn’t getting driven right. It turns out that the secondary of the transformer measured 20k ohms. Probably same thing that happened to the primary.
Today I got out of the hole I was in! it finally works! It took a long time to get all the bugs out, especially since I was adding problems as I went. The clipping was either done by the capacitor, or its bleed resistors. I wouldn’t have guessed that was it, because a few days ago I was measuring 315vac on it. I replaced the capacitor and it worked. I got everything working with no load, and 50 ohms between the capacitor and the IGBTs. There was almost no drop, so I expect it will work with no resistance.
Today I did some more testing. I had a weird problem, there was a short between the hot and neutral, so my current limiting resistors started smoking (1/4 watt resistors pushing 10 watts each). I removed different sections until it worked, then I put it back together because the problem “went away”. It now works with no current limiting, anywhere in the circuit. It does pull 60 ma with no load, though.
The 60 ma is ok. Justin Hays told me it was probably the c-e capacitance of the igbts charging and discharging. After a few rough calculations, the numbers made sense (thanks Justin!). I reformed the capacitors today. One of the electrolytic capacitors was shorted, and another one was accidentally soldered together. The capacitors hold charge well now.
Today I just did some work on bolting stuff together
Today I set up the first test with any load. I set it up so that every 18 seconds the microcontroller would pulse it with 2 cycles. The test didn’t work! The capacitors didn’t charge. They are shorting to the heatsink again!!! I thought I was done with that set of problems! I put two layers of polyurethane over it! I knew that problem was going to come back to haunt me.
I figured out what the problem was! A wire connected to the negative part on the bridge was touching the side of the heatsink. That wasn’t the problem though. Two of the diodes on the bridge somehow failed. They failed completely shorted, instead of a low voltage drop. The different failure mode, and the intermittent connection to the heatsink made this a pain to figure out.
I set up the test again. This time it worked when the midpoint of the capacitors wasn’t grounded. When I grounded the midpoint, one side stopped charging, and so I unplugged it. When I came back to the capacitors they were all discharged. The increased voltage caused something to fail.
Another 400volt diode died, that’s why the capacitor only charged on one side. Two IGBTs on the same leg also died! That’s why the whole thing failed.
I got some pots to use as dividers, and redid the test with 20 volts in. none of the points look like they should. The midpoint looks like the gate drive waveform did. The top is getting yanked around a lot, and there are 200volt spikes! I said IGBTs are bulletproof, but maybe 3000volt spikes could be considered “armor piercing bullets” ;-).
I twisted the wires coming from the electrolytics, and it helped a lot, but it still isn’t good. I ran it up to 50 volts and the cap is charging fine. I also soldered some caps right next to the IGBTs.
I killed the 100k pot I was using. It burned up because of too much current. I replaced it with some ½ watt fixed ohm resistors I got some MOVs to protect my IGBTs and diodes. I looked more closely at where the stray inductance is. Some was in the electrolytic capacitors, it got better when I put a 40uf motor run capacitor right next to the electrolytics. It looks like half was between the electrolytic capacitors and the heatsink, even though the wires are twisted. The other half must be in the heatsink. I closed the biggest current loop I could find, but it made no difference on the waveform!
I didn’t have much time to work on it today. I only tested the motor run capacitors, I thought one of them wasn’t working right but it was. I killed a 10 ohm ¼ watt resistor doing this. I was using it as my fuse/current limit but forgot to check how much power it would dissipate! 40 watts is more than it could take ;-)
Today I had it running at 280 volts, pulsing the primary coil and capacitors for 2 cycles every second. But now it is dead. First I added two 40uf motor run capacitors as close as possible to the H-bridge. The noise on the supply almost went away, that is good. I then made a resistor divider with half-watt resistors, and ran it to 70 volts. I then used a 10k pot in series, but no divider. I got impatient waiting for it to charge, so I turned the knob (or shaft, I don’t have a knob for it) and then I heard a 60hz buzz as the resistor burned open. I decided to just use the three 50 ohm power resistors I already had in there in series. It was working fine, and the capacitors were making clicking noises. At around 280 volts, the microcontroller started freaking out every few cycles and sometimes shutting off. I am almost certain it is picked up from the primary coil a few inches away. I modified the program to flash an led every cycle so I would know it is working. I then looked at the power to the microprocessor. It went from 0 to six volts, following the 60khz current in the primary. I never put any decoupling capacitor on the micro. That was stupid, as it could have saved my project. I will put some fat ones on before running it again ;-). I was just about to turn the thing off, when I heard a loud pop. After I turned it off, I looked over and saw smoke. The pop was loud enough to make my dog go hide. If I assume the voltage across the capacitor bank was 280 volts, then there would be 88 joules in there. The pop killed seven out of the eight IGBTs that I was using! These things are tough, but not indestructible. I can find ways to break anything ;-). I don’t know what caused the smoke and the noise. The IGBTs are all in one piece, and have every pin shorted together. My guess is that the microcontroller got confused and told all of the IGBTs to go on at the same time. I might try some brick IGBTs, because the thing was pretty much destroyed, only the diodes are on the heatsink.
I got the replacement igbts today. I had some clean up to do before I could work anymore so I only got the voltage clamps on the gate.
I got the igbts back on the heatsink
I had more bad connection problems with the 9 volt batteries. I also have more shorts to the heatsink!
I got the igbt bricks today! The other setup was getting really frustrating so I am going to try to use these. I tore up a broken monitor we had for a toroidal core. I wound the transformer today. I am going to try to use 1 transformer for the whole bridge.
I think that because this transformer is on a toroidal core, the windings are twisted as much as possible, and I am not going to put too many turns on this time, it will work with one transformer.
I measured the leakage inductance of the transformer it is around one uh (seen from the primary, with one secondary shorted). I tore the gate driver board, and the microprocessor off the heatsink-spool assembly. I changed the program to skip the whole deal at the end, because I thought it wasn’t needed. So far, the gate voltage only goes up to 20 volts (instead of 30), and it is very slow at doing that. After the “chirp” there is a lot of ringing >8( . now the gate drive chips are getting hot enough to melt hot glue on them, with no load. I killed two pins on the microprocessor, both were the signal for the same gate drive chip.]
I got a lot done today. I set it up so I had 1 brick switching 12 volts. The gate waveform looked fine, with almost no ringing.
I tried to add the other brick, and now the gate takes a while to reach 30 volts, like an rc thing. I killed another processor and gate driver chip.
I couldn’t find 5 volt zeners at either radio shack I tried! I tried to use LEDs for zeners, because they usually have a higher on voltage. I glued some protoboard next to the micro, so my capacitor isn’t just sitting in the air. The LEDs pull the microcontroller’s signal too far down for the gate drivers to work. One LED died, and so did the pin driving it.
I got some zeners, and replaced the LEDs with zeners, and got all of that stuff fixed up.
I tested the bricks up to 104 volts, but with no load, it worked fine.
I did some low voltage testing with the primary coil and capacitors hooked up. I got a weird looking waveform, it wasn’t quite square, it was high, then dropped a little, then went negative, raised a little… It was just because it was out of tune, and was pumping current through the diodes. I only had a few electro lytic capacitors hooked up, so the voltage on the IGBTs wiggles around. I did a lot of rewiring to use all of the electrolytics. I also placed them inside, and wired up some diodes on a little heatsink.
I had shorts on this heatsink, it was the burrs on the heatsink again. I started running the tests, but I realized the oscilloscope was kinda freaking out, there was noise on the scope even when no probes were attached, and it was set on “ground”. I turned it on its side, and it got better, but I still see noise when I touch the ground to the micro’s ground. I tried using both channels, and using the subtract function. The noise went down, but was still there. I “think” there may be some noise on the microcontrollers power, but it’s tough to tell. I cranked it up to 80 volts, at 100bps, and 6 cycles. I had 100 ohms before the rectifier. The voltage from the variac was 112, and after the resistor was 39, so the total power dissipated in my system is 28 watts. The total power then, would have been about 200 hundred watts (probably within a factor of 2), had the secondary been in place.
I tried making a back-terminated scope probe that Steve suggested, and moved the scope further away. It helped, but I still see noise.
I put a 20-volt zener across the DC supply (at the IGBT) to find out if the ringing is real. It’s not!
I changed the micro’s clock from the crystal to an external potentiometer, wound a new primary, and glued the bottom of the secondary back on.
I decided that today, it will make sparks. I just hope they come from the toroid ;-). I chose a tap point by using the function generator, and twisted the pot. To match the freq. For the ground I used a little bit of chicken wire, and a ground stake that is around 2 feet deep. I think I need a better ground. I turned it on, with 50 volts coming from the variac, and got ~2 inches spark. It was running at 10bps. When I was cranking up the voltage, I saw some smoke. I said “Oh darn!” (or something to that effect ;-)), and turned it off. It still works, and everything was cold, so it was probably either dust, or my imagination. I turned up the voltage on the microcontroller to speed the clock up. With the right tuning the sparks jumped out to about 6 inches, with the variac putting out 34 volts. I was busy playing with the arcs, and all of a sudden, it stopped working so well. The arcs would only jump about 2 inches. Nothing looked dead, and I was still in tune. I have no idea what it could be. I used some wire wrapping wire to short the bridge when I was testing the resonant frequency, but forgot to take it off, so it blew open when I turned it on :-).
Last night I was going to show off my incredibly huge 6-inch arcs, but my coil wouldn’t work. Basically, what killed it is that my dad had borrowed my micro’s PSU, and left it on 12 volts =8O.
It took me a long time to figure out what was wrong, because I thought my zeners would have saved it. It took my all day working on it off and on, to figure it out.
My first thought was that my ground was loose, or my secondary was open. Even after tinning the ends of the secondary wire, I still had a flaky connection. I opened my secondary up where I had to solder the wire back together. After resoldering it, that problem was gone, but it still wouldn’t work :-(.
I then thought, “Maybe its way out of tune!” The secondary resonates at 57 kHz, and my primary resonates at 74! I moved it up a turn, and the freq. dropped to 45 kHz! I moved the tap back to the original place, and it read 54 kHz. Hmm… I can’t figure that one out. Still don’t f***ing work!
I finally decided that my IGBTs weren’t doing their job. The voltage wasn’t very square, but I didn’t know what I could trust. The only measurement I could trust is using the oscilloscope with no probes to measure magnetic field (anyone know how to calibrate my new current sensor? ;-)). The gate voltage looked messed up, it looked like an RC time constant that took along time. The output of the gate driver chip oscillated at several MHz when I tried to look at it. When I came back to reexamine it, it wasn’t square, but it didn’t oscillate. The gate voltage also changed to be spiky. I then looked at the microprocessors output signal; one was dead. Doh!!! I know what to do tomorrow, but I still didn’t figure out a lot of what was going on.
It wasn’t actually the microcontroller that died, it was the zener protecting it. I replaced the zener with a 3.9 volt zener, and then that pin did die. I had some more gate driver chip problems, and killed 7 (!) trying to figure it out :-(. I killed 4 of them because I tried “double stacking” them to decrease the resistance. I took out one of three dead time counts, and I think I am ready to go again.
More sparks!!! I did some stupid stuff like plugging in my new variac, and then plugging in the coil upstream of the variac. I didn’t know at first, because it was so out of tune. I was playing with the micro’s clock speed (tuning it) and I was about to jump up to 4.5 volts to get the extra range. Just when I put some pressure on the selector, the coil burst out in ~16 inch arcs! At that point I also had a few turns on the secondary shorted. I turned it off right after that. At first I assumed it was the shorted turns bringing it into tune, but it wasn’t it. I was doing some low voltage tests, and was about to move up to 4.5 volts, and then a 3 inch arc jumped out (instead of 0.75 inch arcs).
I did some pspice today, and realized my tuning was too far off. The way my primary is set up, I can only tune in 1 turn increments, and that’s not close enough. I am going to add an old copper primary in series, for an off-axis inductor. The micro’s ground fell off again, and I first thought the caps weren’t charging. I made some bad measurements, so I took the thing apart, and realized that part worked. Doh!!!
I put it all together today. Acid core flux is great for one time use, but it makes it impossible to resolder >8(. I just have to tune it before I run it.
I changed my off axis inductor for another one for more tuning range, and then tuned it. I ran it with 50 ohms in series, and it worked fine all the way up on the variac. I took the resistor off, and then the connection on my diodes broke again. When I was fixing it, I got zapped. There was no voltage in the capacitors, and it was turned off, but the workbench interrupts the neutral not the hot. My dad had to rewire it once, and made a dumb mistake. It’s fixed now, but it has already gotten me four times (!) and my dad twice. I finally got it up and running again. It worked, but it sparked just about the same at 60 volts as full power. I turned it off, but when I turned it back on to show my family, it went boom. It worked at 30 volts, but then I turned the variac up all the way and it exploded, sending IGBT pieces across the car hold (garage). The voltage clamps on the IGBT work, and so does one pin on the micro. Tomorrow I will figure out what’s dead and what’s alive.
Everything else is alive. It exploded because I was pushing too much current, and the IGBTs overheated, while the case was still cold. Got to test the current, and fix the gate drive circuit.
I worked on the current test today.
At 1500 amps, the voltage drop is around 8 volts. That should be more than enough for some good sparks, so my gate drive stuff must suck.
Here’s a picture of the test setup hot-streamer.com/chunkyboy86/igbtcurrentest2.JPG . I used some 30 gauge wire for a current limiting resistor, and some more as my current sensing resistor. The gates were driven with 4 9-volt batteries in series.
It’s been a while since I wrote in here so I will see what I can remember. I found some 14 amp, 25 volt gate drive chips on digikey. They had the same pin configuration, so they should just be a drop in replacement. With the higher voltage, I thought I could get away with using the same transformer. I bought a 16 vac transformer to power the gate drive chips. The 14 amp drivers had problems with high frequency oscillations. Even with a lot of decoupling capacitors mounted really close, I couldn’t get a clean square wave out (or in). The protoboard it was built on was really bugging me with the connection problems, so I built the circuit on a socket for some computer thing. I just wired the pins on the bottom the way I wanted. I tried using a transistor and pull up resistor to get the input voltage above the oscillating range quickly, hoping that would solve my problem. It might have worked, but I killed my last 14 amp driver. Instead of waiting from parts from digikey, I went to fry’s to get some mosfets and try making a discreet gate drive circuit. I used some of my tc4429s to drive that. Because the tc4429s can only take 20 volts, I had to make a new gate drive transformer. I got the core from an inductor/transformer assortment at radioshack. The transformer I took it from was made by coilcraft. The transformer has 5 (or was that 6?) turns of 10 wires stranded together. Two are for the primary (in parralell) and the other 8 form 4 windings of 10 turns. I tried measuring the leakage inductance using my 50 ohm function generator and scope, but it was too small for me to measure. With the output shorted, the voltage on the primary never got even close to half the open circuit voltage. There were some resonances messing me up at the high freq. I tried discharging a capacitor into it and watching the freq. but I had problems triggering the scope on it. I still had connection problems, so I decided to make a PCB. I should have made that decision about a year ago. I finished designing the pcb and ordered it today. The miniboard service from expresspcb gives me 3 boards, so if you want one email me.
I forgot to add that I got some better IGBTs from ebay. They are CM150DU-12F instead of CM150DU-12H. There are some big differences. The gates take about 3 times the charge, but the voltage drop and peak current handling capability is much better. The effective capacitance I am driving is about 1.9 uf!!!
I got the boards today. Right after ordering the boards, I found a mistake. Doh!!! L I emailed them and asked them if they could cancel that order. They said sure, so I fixed it and ordered more. The bad ones showed up too. I built up the three good ones, except the microprocessor, because I can’t solder it. I am surprised I could get the 603s. I used the dremel tool to fix the other three boards. The other two boards are also missing the gate drive transformers.
Here are the problems with the board:
I was off by a factor of two on the gate capacitance; it is effectively 960 nf. My dad was able to tack down the microcontroller so I was able to do some tests. The rise time on the capacitors (gate simulator) was terrible. It took around 2.4 us. It was less than critically damped, but there was a considerable amount of resistance too. Judging from pspice sims, I have about 0.5 uh, and 800milliohms. The MOSFETs are 50 milliohms for the n-channel and 113 or so for the p-channel. I was getting weird delays, which resulted in unwanted dead time. The problem was that the pull up resistors had too much resistance. 560 ohms was too much also. I found some more problems with the board.
I removed the transistor-pullup resistor stage, and it works better. Instead of removing the pull up resistor, I shorted it (What was I thinking?). The microprocessor survived, but one of the p-channel fets didn’t (how’d that happen?). After adding decoupling capacitors (two 10uf and 1 1 uf to each half bridge)to the MOSFETs, the rise time on 880nf (no transformer) was around 960ns. With the transformer, and 340nf, it took around 3 us. I rewound the gate drive transformer with a lot more copper. The primary was has 6 turns of 15 30awg and 10 26awg in parallel. The secondary wires are 22 gauge, and there are two per secondary (series). I think I messed up the seriesing (it is a word now ;-)) of the secondaries, because the thing is shorted. I think I killed an n-channel fet now L.
The n-channel fet was fine, the zeners died though. I spent a lot of time chasing the problem of a shorted transformer. It turned out it was that one of the secondary traces was bridged to the other one. I also had the problem that the primary of the gdt was connected to the ground plane. That was the case of the pot touching the primary wires. It poked through one layer of electrical tape, so I had to add another. I got ok rise times at the other side of the transformer and long twisted wires (8”?). It took 440ns to reach 20 volts, and about 900 to reach 40. The ringing went up to 60 volts when unclamped. I have to find a metal container/shield for my pcb. The heatsinks are supposed to be on there way. I also have to test the gate voltage, to make sure it can stand 40 volts. The gate capacitance is more, the gate leakage is more, and lower gate voltage keeps it saturated better than the “h” ones, so I was thinking it might have a lower oxide thickness. It does have the same threshold voltage, and is rated for the same voltage, so we’ll see if it holds. I had a perfect power supply for this test. It went to 42 volts, which is perfect. I decided to charge a 32000uf 25 volt capacitor to 40 volts ;-). Somehow the capacitor survived, but my power supply didn’t L. Now I have to find a new way to test it.
The 3 out of the 5 IGBTs were completely dead, and the other two were half dead. The guy said he would send me 7 more. I tested the gate of one of the half dead ones. It failed at 45.6v. the test setup was 7 9volt batteries, with 1k limiting resistor driving 1000uf in parallel with the gate. I have been having a discussion with the sci.electronics.design group. They all say I need a bigger IGBT. The rise time is good, but the TVS doesn’t clamp until 49 volts. The rise time to 20v is 300 ns to 40v is 450ns and to 49 volts is 580ns. This is all with 440nf on the secondary side instead of the “real” 240nf. I need to do a current test with an inductive current limit. I want to add on to the current/vce drop plot that the datasheets have. I may want to go with a lower drive voltage to add some margin for error. I found the transient thermal impedence chart on the datasheet. The thermal capacitance of the h-bridge is 0.11 joules/degree C. That means, with a case temp of 50 C I can do 40 joule bursts 500 times per second, and keep the junction temp below 120 C.
The rest of the guy’s IGBTs were bad too L. I got the whole thing built up in an aluminum box. I had some problems driving the gates when they had TVS on them. When I lowered the power supply voltage, the voltage on the mosfet gates got lower too. I forgot to change the zeners. That was causing me some problems. When driving 440 nanofarads, you can feel the clicks every pulse! That’s cool ;-). When driving one gate with 28 volts TVS on it (really 33v), the rise time to the peak is under 100 nano seconds. I did some more pspice simulations with guessed at heatsink thermal resistance, and calculated thermal capacitance values. The results look promising J
The microcontroller doesn’t always start on power up. The power led will light up, but the microprocessor doesn’t light up until it gets turned off. It obviously stops quickly because of lack of power. If I turn it on right away, then it works. I can get some good gate waveforms on the gate now, and will post them in the pictures section. I now have to figure out how much dead time I need.
I tried adding some dead time because I thought it was needed. At this clock frequency, the resolution is 2us, that’s no good. In that two microseconds, the gate voltage goes from –35 to 20 and back to –20. I did some more tests, and at it only pulls 16 ua (15 cycles, 10bps, 9v, 1 brick)
I forgot to mention that the transformer would start to saturate at 58khz. I realized that the code called for 8 counts up and 7 counts down. No wonder it was saturation on one side only J. It’s all better now, and wont saturate at the lowest freq. I’ll check to see what that actually is.
Stuff to do
I had extra time today because there was no wrestling practice J. I wrote the extra code, so now the other potentiometer adjusts break rate. The break rate is limited to 50 bps, because above that the gate voltage starts to sag due to a weak power supply. I had problems with the register overflowing, and it wouldn’t work as a long integer. I then had to learn how to do nested commands, which was kind of a pain. The limiting the break rate part was actually easy. I also had to add a capacitor at the input to reduce the noise. It turns out that I don’t need an audio pot. The audio taper actually made it too sensitive in the opposite direction. I just fixed it in software. When the register overflows, the break rate usually changes to really slow. I noticed that one time it went up to 1000 or so. I can just imagine playing around with it, and having the register overflow, streamers connect to me, and then an explosion of the IGBTS. I wired it up for the next test, but it’s too late.
I tested it at 200 volts and no load. It worked. I bolted the IGBTs to the heatsinks and set the coil back up. I tried to get it all tuned, but the coupling is too low (~0.3). I am going to get some copper tube to make a real primary. I decided to copy Greg Hunter’s primary. I cut the pvc pipe, and drilled the holes.
EDIT: coupling was 0.03, 0.3 is high!
I got the primary built today, but no time to run it :-((((((((
This thing is tough to tune! It may be because the microcontroller is slightly asymmetrical, but I couldn’t get it to tune right. Even without tuning I managed quarter inch sparks when powered by a nine-volt battery ;-). The micro’s output is asymmetrical by one clock cycle. I can’t fix it now because the laptop is getting fixed >:-((. The asymmetry was also causing saturation in the gate drive waveforms. I didn’t notice it because it was only the last few cycles. Pspice says that the asymmetry doesn’t matter for tuning, I wonder what my problem is.
Pspice gives some cool facts to help me tune it. If the primary freq. is off, The wave form still looks right. If the primary resonates too fast, the switching times will all be too late, and if it’s too slow, it will switch too early. If the driving frequency is off by a little bit, the waveform gets messed up alot. For even finer tuning, the switching time helps again. If the driving frequency is too high, the switching times will be too early at first, and then too late. The opposite, of course, is also true.
With the nine-volt battery power, the voltage across one capacitor is 25 volts. That means 50 volts across the whole thing, and 3 millijoules transferred to the secondary. At 50 bps, it was using 0.15 watts :-O. I hope it will handle more J
I got the laptop back today and fixed the code. The output of the IGBT doesn’t look so square, but I think it’s just the 2 volt drop. I’ll see when I get to higher voltage. The voltage on the IGBT itself gets pulled around at each switching time. It is either shoot through, not real, or something else. I don’t think it is shoot through, because it passed the other test and barely leaked any current. If I ground it with channel 2, it looks worse than if I use the right ground. I have to check what it says if I attach the probe to the ground spot. Other than that, I think it’s ready to go!
I asked Steve C. about the ringing thing. He said it was probably the diode recovering. I changed the tuning and unhooked the load. That didn’t fix it. I am back to thinking it’s shoot through. It is around 7-8 megahertz. When I have a nine-volt battery next to it, it only lasts 100ns. When there is no 9-volt battery, it rings for longer. I decided to go to higher voltage to see if it went away. It didn’t L since it is so little energy, the IGBT might be able to absorb it. I didn’t see any mention of avalanche energy on the datasheet, but I hope that’s because it is in dual configuration, and not that it works differently.
I ran it up to 40 volts dc (14vac) and made some sparks! It was probably around 6 inches, but that is a rough guess.
When it is running at such low power, the tuning changes a lot as I bring the grounded rod closer. I got out the old scope and my differential probe to look at the primary capacitor voltage while it was running. When it arcs to ground, the primary voltage and current build up a lot. I have to figure out how to prevent that in high power tests. My plan is to reprogram the microprocessor to monitor the voltage on the primary capacitor (through a resistor divider + a capacitor for a low pass filter) and shut off if it reaches a certain level. It can’t shut off immediately, but will finish the current cycle first. It should also shut off if the slope becomes positive for a second time.
I didn’t get much work done this weekend :^(. I had a wrestling tournament yesterday, and was too tired today. Last night I realized that the stupid choke idea wouldn’t work, because there is no way to have inductance in the shoot through way, but not normally (without huge diodes).I put a 10 volt zener across the IGBT, and it clipped the ringing, meaning it is real, and fixable ;-). It is only shorted for ~50ns, so the only problem I have with it is the overvoltage. I put the 400v transient voltage suppressors back on, and went up to 57vdc. I have to get the second channel (2000x instead of 10x) of my differential probe working before I go up anymore, because I want to make sure it is clamping. I cut down on the amount of cycles, so now the current is limited to pi times what is should be. Now it is easier to tune, because in tune produces biggest sparks. I have to remember to turn the breakrate up to 100 or so, now that I am using less driving cycles. Before turning it on all the way, I have to make a voltage divider, and reprogram the microcontroller to shut the current off if it gets too high.
I fixed my differential probe, and ran it up to 108 volts dc. I added an extra microfarad across each half bridge. I don’t know if it helped yet, because I haven’t looked at it with the good scope and probes yet. The sparks were about 6 inches still. The volts/inch is staying linear, so I am not being stupid again. I don’t see any sign of current limiting, and the heatsinks stay cold. The square wave out of the IGBTS isn’t too square. It takes a long time to switch, and there is a bunch of high frequency hash in between. I think it might be my differential probe, because I haven’t added the capacitor to compensate it yet. The first time I turned it on the primary tap point was in the wrong spot. I fixed it and the sparks got bigger. I may still be off the optimum spot. The differential probe started freaking out. The whole screen of the scope lit up with a high frequency sine wave. When I unhooked it and shorted it to itself, a weird semi-squarewave showed up. I brought it in, and after working for a second, the problems were duplicated on the good scope. When I was drawing arcs to ground, the scope’s screen went completely blank, and I couldn’t find the trace. My guess is that tomorrow when I check the probe again, it will work.
There is no second beat, which is weird. The energy appears to stay in the secondary. When I look at the output of the IGBTs after they’re done switching, I see a sine wave that is about the amplitude of the switching voltage. I wonder what is going on there. Maybe tomorrow…
My probe works again! I figured out why the waveform was decaying so much faster than it was rising, and why the energy was staying in the secondary. One of the gate wires fell off, so I was driving it with a half bridge.
I got it working again, and it works better now ;-). It made sparks somewhere around 12” long at 100 vdc input. I still need capacicitors for compensation on my differential probes so I can make sure the ringing is clamped. I did a lot programming work, but I still haven’t gotten the overcurrent protection to work yet. I decided to not wait for the differential probe. I will just hope the zeners and MOVs work.
I shorted the secondary to simulate a worst-case scenario, and turned it up to 200v. It worked perfectly ;-). I chickened out and stopped there L. According to pspice (very useful tool!) that should be around 1360 amps. I unshorted the secondary, and turned the power up a bit, and it made some sparks up to ~~16” long. This was at 200vdc (72vac) and a coupling of 0.12 (8 cycles). This calculates to 5.2 joules 460kv, and 465 amps peak. I ran it up to 130bps. That should be 676 watts. I got some video of this I will put up on my site. The primary was tuned for no streamers, and no grounded rods nearby. This probably threw it out of tune, so I was probably delivering less power than stated. The heatsinks did not heat at all :o))
The variac goes up to 135 volts, and I can lower the coupling, so I can push it quite a bit more. Before it makes huge sparks, I have to set it up so I can control it remotely.
I did some more work on the current limit, and it worked with a dc voltage on the pin, but not out “in the field”. The zener fell off =8^O. Luckily I was running ~0.5v (on that pin) at the time. It would shut off as I changed the clock frequency, but not the input voltage. Before going to full power I will have to fix this. When I run this thing, it fools my dad’s computer into thinking that there is a fax coming in… I am going to add wheels, bolt on the secondary, and make a separate box for control. From now on, high power tests go outside.
I was playing around with it at 80vdc, zappin’ CDs, and bottles of low-pressure steam. Somehow, I got zapped. It hurt!!! :-(( My arm was hurting afterward. I am glad it didn’t get me at 200vdc. It is probably good because it gave me respect for it. It scares me now.
I retuned the primary, but it didn’t make much difference. I tuned it by looking at the switching times, but didn’t notice any more output voltage on the scope. I was off by about 1/6 of a turn.
I glued some more support things on the secondary. They are 4 layers of Plexiglas, and stick out about 0.5” They hold the secondary on well. I bolted the heatsinks to the spool. They are kind of loose, but it works. I also put wheels on it so I can move it to the driveway for the next tests. I didn’t get it centered too well, so I have to add a fourth wheel. I still have to chop off the screws pointing out, solder the extra motor run capacitors on, add a fourth wheel, put the PCB in a new box, and Velcro that box to the heatsink.
I got some coax for the primary current measurement. It turns out the shield is important when there is such a huge electrostatic field. I also need some shielded cable to get from the control box to the other one with the PCB in it. I need 5 conductors in a shield, and I don’t have anything like that now.
I decided to make a current transformer and actually measure the primary current instead of the primary voltage. This will require some change in code, due to the 90-degree phase change. It should be less sensitive to the electrostatic field, because it will be low impedance (z<1ohm). I experimented with using a spool of wire wrapping wire on a flyback core (no gap). I got some pretty good results, and the turns ratio was almost exactly 250:1. I started to set up a max current test, but ran out of time. My secondary broke in the middle a couple days ago. I have to solder it, and reinsulate it. One of the nuts for holding the secondary on won’t go on. I just have to use the grinder thing to make more room for it.
I finished the current transformer. It has 6 feet of coax, and is terminated with 1 ohm at the end. It should be around 4 millivolts per amp. I plan on calibrating it tomorrow by connecting a ten-ohm resistor to the output of one of the IGBT half bridges. I will measure the current with the transformer, and measure the voltage across it at the same time. I haven’t decided if I want to just record the exact v/a or if I want to make it some even number by adding or removing turns (or by adding resistors in parallel)
I tested it to 1700 amps today!!! I used my new current transformer, which is 4.7mv/amp, and got to 8 volts. This was with 8 cycles driving it with a shorted secondary. It got to 123.6 vac in ;-). I recorded the results, and went outside to push it harder, but I got some weird results. Right above 8v (out of the current transformer), there was a high frequency 2.5 volt signal that was also triggering the scope. I didn’t know what was causing it so I stopped. My dad was outside (I was inside, the garage door was a blast shield) watching for secondary arcs and stuff, and he said he heard it “change beats” or something. 1700 amps is good enough. I think I won’t push it any further for now.
I have to resolder the diode bridge now :-((. The test I did was with a newer current transformer. I wound two windings of 100 turns (28 awg magnet wire), and put them on opposite sides of the core. That way the primary coils magnetic field won’t interfere like it did on the previous transformer. It works better.
I tested the current transformer by using my IGBTs to drive a ten ohm resistor. I measured the voltage out of the current transformer, and voltage across the resistor. I came up with 4.7037 mv/amp. It is probably a few too many significant figures though…
I decided to make a peak hold type circuit, so I don’t have to measure it at any specific time. I finally soldered on the extra capacitors. I also soldered on the diode again, and resoldered the gate connections that fell off, although I am not sure I got the right polarity. The little butane powered soldering iron puts out a lot of heat, and soldering 10 awg was no problem.
Big sparks!!!!!!! It struck the jet ski, which was 47 inches away, and probably hit further when it hit the target in a previous test. I should have measured the distance before moving it. I am not sure how close to being in tune it was. The primary LC could be off a little, and the waveform would still look right. The drive frequency could have been wrong because the streamers detune it. I was too excited/stupid to check for the second part. Assuming it was in tune, it could have been putting out as much as 13.7 joules and 1780 watts at around 130 bps. The input voltage was 117.5 vac. The peak current was 736 amps (calculated). I can easily lower the coupling for more energy, and the IGBTs won’t overheat until really high power. At 1700 amps, I can get over 70 joules! I don’t think I am pushing it very hard yet. I took some movies of it, and will try to get my brother to get them on the computer for me. I was using a breakout point. Streamers formed fine ;-)))))
I had it running at about 10 vac, just to tune it initially. I left it running while I set up the target. I thought to myself, “hmm… I wonder how far away it is.” So I measured it with my hand. Then I thought “I am an idiot”, because a 2 inch arc got my finger :-((.
If I use the 220k + 1pf/foot model, then the driving frequency would be off by about 1khz, causing more current flow, but about the same energy transferred.
I just realized that the simulation says I should get 2400 amps, and my current transformer said 1700. I can see more reasons for the simulation to be wrong than my transformer, because I think I calibrated it pretty well. If the real current is only 1700/2400 times what the simulation says, then I would have been transferring 7 joules, and about 900 watts. The whole thing was probably out of tune, and it probably arced further to the target than the jet ski, because the break out point was pointed away from the jet ski. I will have to do a better test tonight.
I looked at the movie frame by frame. I found that it struck the jet ski several times, and the 47 inch strike was the shortest. I had already moved the tesla coil, so I can’t get an exact length. By going 47” away from the first spot on the trailer in the direction the tesla coil was, and then measuring to the new strike spot, I measured in the ball park of 66”. I moved the jet skis so I will have more room for the next test.
I fired it up again. The most I got was 61” at 108vac. If I turned it up any higher. I get racing arcs :-(((. There were 2.5 volts coming out of the current transformer meaning it was about 530 amps. I wasn’t able to get a good notch, so the primary average VA would be more than normal, but it wasn’t all transferred, so I will assume it cancels out J. It calculates to 9.1 joules in, at 125 bps, or 1140 watts. I have to put the Plexiglas rings around the secondary before I do any higher power testing. I tried it without the breakout point, but there were racing arcs before it broke out. The heat sink stayed cool the whole time. I need to get a wattmeter
My dad brought home some shielded cable for me, so I can get put it into two boxes tomorrow. Today I put 3 more coats of polyurethane on it, and ground out the centers of the Plexiglas discs, so they would fit. There are 6 discs, and they add about 2.75 inches to the radius. This should more than double the creepage length, which will hopefully fix it. Real time tuning might also help.
Yesterday, I put 7 discs on the secondary. The top ones are on a little crooked L but it doesn’t matter really. I also got the cable attached to both boxes, and all the solder connections are all done. The driver part works.
I did some more pspice simulations, and I might have been using an extra 10% more power than I thought.
Running the microprocessor clock resistor 30 feet away from the microprocessor turned out to be a problem. Who would have expected? :-P
It is a strange problem though. The clock frequency wont stay still at low frequency, but is stable at high frequency. There are certain points at which it goes crazy, but on either side is fine. By slightly turning the pot, I can change the beating frequency.
I changed the code to just put out a constant square wave of 1/3 clock frequency. I saw no more waving around. I’m confused.
The problem was that the ac power was run through the same shielded cable. I think it was capacitively coupling to the other wire, and messing it up. I twisted a pair of 28-gauge magnet wire, and ran it next to the big cable. That fixed it, but when I turn the LEDs back on, the problem comes back. Turning the LED on 1500 clock cycles before the burst lets all the transients settle down.
I am an idiot. I blew up another IGBT, but barely. This time only one die went, and the case is still intact. When I turned it on, the current waveform and everything looked fine, but possibly a bit low. I turned it up to 30 vac, and didn’t see much streamers. I wasn’t sure if they were supposed to form yet, so I went up to 50 vac. I then realized that if I turned the breakrate down, the current would build up, and the lights would light up. The power supply was being pulled down. At first, I thought there was a high impedance in the line, so I took the line filter out. The variac was making it’s normal noise, and it wouldn’t have if there was no current flowing. Every time it fired, you could see the Christmas lights dimming. I turned it back up to 50 v, because I was being a retard. Then I heard a pop :o(((( the heatsink was warm. The IGBTs pretty much had to be shorting during the burst, and had to be alive until the pop. My first thought was that the gates were wired up backwards. I had resoldered the connections today, but each gate had one connection still on it. It turns out that wasn’t the problem. I put the new IGBT on. It was doing the same thing, but this time I only went to 10 vac. I was very frustrated, and it was getting late, so I called it quits. I have to move the current transformer to look for shoot through current.
Assuming 50 vac, and 1700 amps (probably more), then each IGBT die would be eating 20 joules! >:-O
It took me a while to debug, because my current transformer broke. Not only was I missing a reading, but there was some added inductance in series. After fixing that, I got it to where it would wouldn’t pull current, and then kept adding connections until it stopped working. It never did, it works again. I may never know why that IGBT died.
I turned the power up even more, but the sparks are a little smaller L. I turned it up to 12.2 joules per bang, and ~2500 watts. At 600 watts (100bps), I could hit 4 feet, but I didn’t get to 5 feet. Even though 2500 watts was at 230 bps, I would have expected bigger sparks. The current waveform looked good, and the heatsinks were only slightly warm. On the final full voltage run, I lost control of the breakrate!!! It was stuck around 100bps, and I had no control of it. A few minutes after power down, I tried it again, and it worked. The streamers detuned it a bit. I had to change the tap point by half a turn, and on low voltage, the waveforms look ugly.
I had some more problems today. The power was shorting to ground through the PCB box. The bolt I was using to secure the power touched the side. I was afraid of that. The gate wire also fell off. I was able to get some 65” strikes, and some thick arcs at 45”. I got it to breakout with no breakout point, but at near the max breakrate (230bps) it would cut out, because the line dropped enough to prevent breakout. I am giving up on the current limit, because even at only 45”, the current would only build up to 20% more than normal. I had to detune more than ¾ of a turn (out of 4)
I raised the secondary way up, and got ready for a high power test. Upon turning it on, I realized that the other 6 cycles weren’t showing up on the beat, and that the sparks were not any longer. Hmmm…. Then I tried taking the secondary ground wire off. The waveforms look the same. The whole time I have been running with a very terrible ground! That would explain why high coupling was doing so good compared to low coupling. What I don’t understand is that I was able to get 5 foot sparks with almost no ground. I did notice that the optimum tuning point was way less inductance than it should have been.
I made a better ground plane with a bunch of aluminum foil. It did pretty much the same thing. I spent some time trying to tune the thing, using tricks learned from pspice, but I couldn’t get a good notch. I got the function generator out, and tuned it that way. Still no notch! I played around with pspice to figure out what could be causing it. If I had 0.2 ohms in series, then it would mess it up like that. Using the great information from Gary Lau’s ac resistance measurements, I calculated around 0.1 to 0.15 ohms from the primary plus the 10 gauge wires I had running around. That is close enough. When running it at low coupling, the primary would get pretty warm. The connections weren’t soldered, and got especially hot themselves. I could also feel some warmth coming off all the 10 gauge wires I had running around.
I ran it a little bit at lower coupling, and that worked fine. I tried doing a double burst thing, with one about 200 microseconds after the first. It didn’t really help much, and the second one would not tune the same as the first. Before breakout, the second burst would be bigger, and after breakout it would be the other way around. I have to do a “real” test on this later.
The insulation was scraped off the magnet wire I was using to supply the controller PCB I stepped on it and touched my grounded metal box (ouch!). Then the connection broke completely. I cut up an extension cord and am using that to power it now. The wires going to one of the gates broke again!
For my new primary, I am going to use two layers of 6” 10 mil aluminum flashing. Although aluminums bulk resistivity is significantly worse, it’s greater skin depth almost cancels it out. It is only about 25% worse than copper. The current should be pretty uniform along the thickness, because 10 mills is less than 1 skin depth at 60 kHz, and current can flow on both sides. I plan on just using 3 parallel 10 gauge wires to get from the electrolytics to the IGBTs, and from the IGBTs to the primary. For the primary supports, I am going to use some pieces of acrylic. My friend’s dad has one of those laser things, so I can get them cut professionally.
The magnet wire that I was using to power the gate drive circuitry broke. I cut up an extension cord and am using that instead. Somehow, the power supply for the microprocessor is shorted. I checked one of the pots, but that isn’t it. The wires fell off one of the gates again.
I got the new primary wound to 4 turns. One of the bottom plastic holders broke, because it is only 1/8th of an inch wide. I tried testing the inductance of it using the function generator to produce a 150khz sine wave. I then measured the voltage and current across the primary. I got 1.4 uh, and it doesn’t change when I short it. I can’t find any place where it is shorting. This should be a very simple problem, but I am tired and sick, so I can’t think too straight now.
I wound the primary, and tested the inductance. I got 14.35uh (+-10%). There wasn’t any problem with it before. I was thinking it was 4 turns but it was only 3, and the tallness reduced the inductance more than I would have thought. I used javatc to compare a tall solenoid to a short one. I got a 2.5x reduction. The measurements and calculations agreed within 6% J. I guess I wasn’t actually shorting it when I thought I was. I broke another primary support, but I still had enough. I also cut myself several times before I realized that I was bleeding, and a few more times before I figured out how I was cutting myself. For tuning, I was going to add one more turn of 3/8th inch copper tubing, or a couple ¼ inch ones so I could tap easily. The turn would be going the wrong way for low power stuff, and the right way for high power stuff (streamer capacitance needs more primary inductance to tune).
I spent some more time cutting up the board looking for the short, but it went away. I had some more problems with the microprocessor not running, and the power being pulled down to 1.5 volts. I decided to power it with a 9-volt battery before the regulator, but accidentally put it after. Doh! Need a new micro controller L. My dad tried to get the dead one off the board, but messed up the traces a little bit. Luckily, I am not using most of them. I am going to see if Margaret can get it soldered together.
I glued the primary together the rest of the way. I also got some ½ inch copper tube to use for the last turn to tap off of.
Margaret got the first board back to me last night. Some of the pins weren’t tacked down right, but I got it working again. I took it outside to hook the gate wire up again. Those darn things keep falling off. One of the gates wasn’t being charged (no, I mean another one that did have wires on it). I was staring at the circuit board wondering where the problem was and then boom! The gate drive transformer blew up. The only trace that was messed up was one going from the secondary to the IGBTs. The primary of the GDT was melted open in a few spots. I am surprised that the primary could get that messed up, and have no obvious current return path. The primary of that transformer had 10x the amount of copper that any of the wires bringing power to the board had. I took the transformer off the board. It works without the transformer. I think I may have a flaky connection though. The transformer is shorted now, and burnt up looking. It was a pain twisting all of those wires together. I am going to see if my uncle has any shielded cable that will work for me.
I redid the transformer with 8 strands of 24 gauge for the secondaries. There are 2 windings in series for each IGBT. The primary is aluminum foil wrapped around the other windings. It was all wrapped with electrical tape. I took some strands from my 10awg wire, and wrapped them tightly around the foil. I then soldered them to keep them tight, and used that as my connection point.
I had another intermittent short. Some solder on one trace was bulging out and sometimes touching the ground plane. I got it all ready to go, but when I turned it on in the garage to check the gate waveforms, it shocked me. The case somehow got tied to the hot wire, and the ground was only 30 awg (stupid!), so it burned out the ground, then zapped me. The switch broke from the high current. It is almost fixed now, and has a thick ground cable. I might fuse the input, that would be a good idea.
The other boards now have the micros on them. They just need gate drive transformers, and some decoupling caps.
I wired the power through 30 awg, and the ground through 14 gauge this time. It all works again. The gate waveforms are about the same as before. It takes ~70ns for the gate voltage to fall from 20v to ground, and around 90ns to get to 20 from ground.
I redid the wiring from the IGBTs to the capacitors to the primary. I used 3 10 awg wires for it. I still have to chop off excess wiring, and attach the ring terminals to the ends. The secondary needs to move even higher now, so I got some threaded rod, but I haven’t put it on yet.
Radioshack doesn’t carry the inductor assortment anymore L. That’s where I got my gate drive core.
I got the inductors free from coilcraft J I also got the decoupling capacitors on the other boards.
I redid all the wiring with 3 10awg wires. I also added a turn of 0.5 inch copper pipe. I tried to test the peak current the IGBTs could do, but I was getting turn to turn arcing. I tried to measure the resistance by letting it build up to steady state and measuring the current. The capacitors discharged before steady state, but my guess is 0.16 ohms. The aluminum got warm, but nothing else did. I tried running it, but the coupling was pretty low, and the resistance wasn’t low enough for it to work well. Tomorrow, I will check the DC and AC resistance and try to figure out what I did wrong. One of my guesses is that it wasn’t pure aluminum. I have to take out the inner turn to tune it, and fit the secondary inside so I can get higher coupling. I noticed that at the wattmeter (one extension cord from the wall), I could pull the voltage down from 118vac to 100vac :-0
The DC resistance of the aluminum was 90% higher than I calculated. I measured about 4.9 milliohms, and calculated 2.6. I measured the dc resistance of 1 foot of 10 awg and came out with 1.05 milliohms, which is close enough to 1 like it should. be. The AC resistance was 27.2v/108.6a, or 0.25 ohms. I measured it by looking at the voltage across the primary when the dI was 0. I put 600watts into the primary to see what part got hot. The inner turn got hot, and the outside turn was cold. The turns in between gradually heated up as you approach the center. I tried using only one layer of aluminum, to see if it was better or worse, but it looked pretty much the same. Time to ask the tesla list!
Bert Hickman told me that it was probably due to proximity effect, and mentioned that most primaries have more loss on the inner turns. I used FEMM to look at the current distribution. The thin winding pretty much eliminates skin effect and current bunching to the inside. The current wants to flow on the top and bottom edges though. FEMM predicted 65 milliohms for one layer and 0.21 ohms for two. The current in the second layer was flowing backwards like I expected, and I guess was still connected when I thought I disconnected it. I used FEMM to predict the loss of a larger diameter primary. 16” ID and same L gives 58 milliohms, so it doesn’t help much, but it is better. The coil fell off of the counter when I was trying to take the turns off. The center piece broke, so now it doesn’t have anything holding it in the right shape. I have it at 15” ID, and am going to use the copper turn to help it keep it’s shape. I just need a form to bend the copper around.
I finished the primary and attached it to the coil. I also re did my current probe with about 20 feet of coax so I can monitor the current while it is running easier. I had to use a different resistor for the current sense, and haven’t calibrated it yet. I messed up on the calculations for the resistance. I was 2x too high. My primary should now be about 29 milliohms.
The micro controller somehow died. It worked last time I touched it, so I don’t know what could have happened. I switched the micro for a new one, and it works again. I had some more bad connection problems too L. I ran some current into it with no secondary, and the waveform looks a lot better. I do see the effects of some resistance though. I tried to measure the primary resistance the same way as last time, but couldn’t measure any J. The coupling would be too low the way it is set up now, so I have to change that before running it. I also have to check the inductance, to make sure I can tune it with the 1 copper turn.
I ran it again at low power, but the TVS fell off, and it blew the other IGBTs. One brick had one device fail short, and the gate and emitter connections for one IGBT were open on the other brick. I made a couple H-bridges with the isotop IGBTs, but haven’t tested them yet. One of the gate zeners failed shorted, so I didn’t have a chance to test it.
I got more gate zeners from digikey.com. I put 28v (33v clamping) TVS on the new H-bridge. I tested the new H-bridge and there is a couple hundred nanoseconds of shoot through. I was going to fix that by adding a capacitor from the input of one gate drive chip to ground. That introduces a delay (RC). That means that instead of switching at the same time, the outputs are both high for a while, and on the next half cycle, both low for a while. This introduces dead time.
I got some more CM150DU-12H IGBTs, so I went back to using those for now. I put the 33v (37v clamping) TVS back on the gates. I realized a problem in my wiring, but it is hard to explain, but I’ll do my best.
Refer to picture 1.JPG, 2.JPG and 3.JPG. 3.JPG is the cross-sectional view of the wires between the IGBTs and electrolytic caps (P1,P2,P3,P4). The transformers represent the inductance of the wiring. When the current is flowing into P1, and doesn’t return out of P2, then there is a fairly large inductance because there is a big loop. When the current flowing into P1 flows back out of P2, the inductance is much smaller, because the loop area of the twisted pairs is small. The same thing goes for P3 and P4. The resistors represent the resistance of the 10 gauge wires.
When wired up as in 1.JPG, all the current has to flow though the two resistors in series, and there is a much larger inductance. When the coil isn’t tuned, there may be hard switching, so the inductance of the wires can put a big voltage spike on the IGBTs. The current might be more than the TVS can handle, so it’s not as safe to wire it this way. The current distribution is not uniform because of skin effect, but it is as close as you’re gonna get.
When wired up as in 2.JPG, The inductance is much smaller. The current also flows in all the resistors at the same time. This may lead you to think that the resistance would be half the first way, however, the current distribution is not near as even, so the resistance of each resistor goes up. This is due to proximity effect. The current wants to flow on the inside because the inductance is lower that way. When the wires are far apart, the change in inductance is smaller so the effect is not as noticeable.
I did not specifically test to see which had lower resistance. I made the changes for safety (for the IGBTs), and the resistance was higher than I remembered it.
I fired the coil up again, but I was having a hard time getting it tuned. I think that the power throughput was almost as high as tuned (maybe 90%), but the IGBTs may be hardswitching, and see more current than needed. The heatsinks were cold, so I didn’t worry about it. I got up to 68”! I was using a paperclip breakout point, and the arcs were to the jack with no point on it. There were 2 or 3 strikes in 30 seconds to a minute (video camera wasn’t on). When I turn up the breakrate all the way (~230bps) it trips the 13 amp breaker. I am not sure how much real power this is because the current is drawn in spikes. My wattmeter was all the way inside too L. Even after 60 second runs at this power, I couldn’t detect any increase in temperature on the big heatsinks. The diodes’ heatsink did get slightly warm.
I ran it for a while with no breakout point. It is pretty cool like that. The arcs usually start on the far right side of the toroid and rotate around ‘til I can see it. Sometimes it breaks out vertically off the ¼ inch threaded rod that holds the toroid on. Running it with no breakout point did cause some problems though. I had a few primary strikes. After the first one, I turned the power down, and was worried. Then I thought “ hey, if it survives one, then it should survive them all right?” Nope L
There were a couple more strikes, and it ran just fine. Then there was another, and it stopped. There was still power to it, but the LEDs weren’t flashing. Here’s my guess at what happened: The strike sent a common mode voltage spike down the line, and caused the 120v mains to arc to a trace on the circuit board. I think that the reason it survived the other strikes is that the power wire was further from the PCB at the time. I haven’t figured out what broke, except it’s fairly obvious that the trace was cut in half and that the microcontroller couldn’t have survived. The 3v zener to protect the micro was also partially blown off the board. I am glad I had a backup board built up ;-). I just dropped the new one in, and it should work now. I also cut a hole in the aluminum box, and have a cable coming out of it so I don’t have to open it up to reprogram it.
I turned it on again today, but didn’t run it up to full power. The new board did not have a capacitor from the a/d input to ground, so the breakrate (and burst length) was unstable. It sounded cool, but I was scared to run it at all. It turned out to be an easy fix J
The problem didn’t go away! The real problem was the programming cable hanging out of the box. I shoved it in and covered it up with foil. It went away.
I got 74” sparks so far! I forgot to retune it because of streamer loading too :-P. The strikes were to the aluminum foil on the wall. I also got strikes to ground.
76” sparks! More if you count the diagonal distance. The closest point on the wall was 76”. The target was aluminum foil on the wall. The coupling was like 0.15 I think. The aluminum primary was sitting on top of some 12oz powerade bottles. With some tuning, the coil had no problem hitting 76”. It did it repeatedly at ~200bps. The whole run lasted less than 55 seconds, (I was messing with tuning voltage and breakrate) and there were 11 different strikes. 8 of the strikes were in a 12 second period .Many of them were connected for several bangs. It even struck a few times at 130bps.
I moved the coil out to 81 inches to see if it could hit it. It could not L. I had another primary strike, despite the strike rail. The streamers went out and curved all the way back to hit the bottom of the primary! It survived this one too J it has survived 4/5 toroid to primary arcs.
At low bps, the secondary flashes over to the primary sometimes. I tried it without a breakout point, and it did it more often. I did see a few streamers forming on top so I turned up the breakrate hoping it would fix it like it does with the breakout point. I got a bunch of secondary to primary (and strike rail) arcs. I then turned it off. I turned it back on at low power to make sure it works. It turns out that I burned a hole in my secondary L. That sucks. The heatsink on the rectifier got slightly warm, the IGBT heatsinks were cold, the primary was cold, the MMC was cold, and the primary clip was a little warm.
I still have to try lower coupling and more cycles. Hopefully at lower coupling I can do without a breakout point.
Ok, I’ll see if I can remember all the other stuff that happened…
The connection to the Christmas lights failed.
I tried running it a couple days ago and it worked until the ground wire came off the secondary (moving the secondary). I had a ton of strike rail to primary arcs. Oops! After that, the breakrate wasn’t consistent. It would wobble around and stuff. I added some more decoupling. I added a 0.22uf and a 22uf capacitor. The 22uf causes a neat delay. Didn’t fix it. At that point I was confused and sought the help of Steve Conner (again) he had a few suggestions and said he was having similar problems. My dad suggested a digital filter “you know, like an digital RC”. I thought “duh, I’ll just add a resistor”. I added a 7k resistor in there too. That fixed it.
I turned that extra copper turn into a strike rail.
79” sparks!!! I just changed the I=8 to I=10 J I tried it at a low bps. There was a loud bang sound and the Christmas lights turned on for a second, but faded out. I ran it with a vertical breakout point too. I have video of it I have to capture.
In the last run, it hit 79” 5 times in 43 seconds of run time. The run ended when the 15 amp breaker blew. The run never made it up to the full 200bps due to current limitations.
I am pretty far behind here, so I’ll see what I can remember. I rewired my hobo 240v socket into a hobo 120v socket on the other side of the breaker. Don’t worry, there’s still a 40a breaker at the back of the house ;-). I tried vertical breakout points, and went up to 14 cycles. I noticed that unless I used a long point, it would stop going vertical, and breakout along the outside of the toroid when at high power. I got 93” at maximum, but my secondary was looking like it couldn’t handle more. I tried to test the maximum current the bricks could handle, and busted one in the process. It wasn’t from overcurrenting the IGBT really… well.. sorta. Anyway, the primary arced over, basically shorting the IGBT to some caps charged to 9kv or so. At some point, noise abruptly appeared, but wasn’t a problem. I got to 1700 amps with no signs that the IGBTs were stressed.
I ran it as a half bridge for a while. It appeared in our totally kickass Spanish video running on a halfbridge. I did some tests of v/spark length, and the results showed a linear trend. Steve’s work showed this also. I ran up to 30 cycles, and it did 5’ at 94vac in there.
I ran it the first day of summer, and the psycho next door finally got mad about the coil. I guess it was causing interference with her tv or something. I saw someone at the edge of the driveway so I turned it off (wouldn’t have if I had known who it was), so she came and yelled at me. As soon as she walked back down, my dad gave me permission to turn it back up ;-). So then she came back up, and started yelling, and my dad told her to back off, and that it wasn’t safe to be there, but she ignored him, so he said it again, to which she responded “if it’s not safe, then I’m going to have to call the police”. It would be safe if you stayed on your own property woman. Hehe. So then a bunch of fire dudes, and a couple cops showed up. They wanted to see it, and thought it was pretty cool. They told me to cool it for tonight, but other than that it’d be ok. Then they went and talked to the demonic hag. It was a pretty exiting day, since I already got in a kinda fight with a good friend, and had a spill of boiling HNO3/H2SO4
Since then, I replaced the brick, but it was a long process getting it back running, to say the least. Sometime late summer I tried running it, but it was shorted. It turns out I had mixed up the positive and negative cables, so when I tied the IGBTs together, I tied them shorted. The rectifier also had a mechanical failure, so I replaced it with an isotop diode dealy, and superglued it to the heatsink.
A couple weeks ago, I changed the code to use the second pot for controlling the number of cycles, and fixing the bps. I also redid the primary, since the old supports were either broken, or had been lost. I also lowered the primary and secondary, but the secondary more, and had higher k. The secondary was about 3” lower than normal with respect to the primary. I tried again, and it didn’t work. The problem was that one of the gate zeners was shorted on one polarity. I took it off, and looked at the output waveform when unloaded. It looked like crap. I also noticed that the GDT was buzzing. I got into a long process of debugging what I thought was a saturating transformer. This included messing with the code to get better symmetry, winding a dummy gdt, and stuff. Somehow I killed the uC doing this, and pulled the regulator off thinking it was dead. It also fell off the table, and shorted mains to ground. Part of the pin got melted, the 3v power led broke, and it got a flaky connection between the pot for the clock, and the uC (I think, haven’t really figured it out)
I switched to my backup board, and wired on a connector so that it could be easily disconnected from the board. It had the same problem. It turns out that my scope was fooling me. The 10x probe showed it working normally, but when I scroll down to get it in view with the 1x, it gets distorted. So the buzzing was normal, and the waveform really was good all along. I tried again, and figured out that the real problem was the zener. Still didn’t work! I had the phasing wrong, so that both sides of the primary were swinging the same way. Fixed that, and it started to work! But then the output went away, and the variac buzzed; I had blown a brick.
The problem was that I changed connectors between the two female ribbon cable connectors, and the pins weren’t quite long enough to connect the one gate. So it failed when the gate was on, and caused shoot through. I got another connector and it worked. I got up to 65” before primary strikes convinced me to stop I also got a nasty primary to secondary arc that climbed up the secondary.
My dad knocked the secondary over, and the primary was sitting on top at the time. This flattened part of the toroid, broke one piece of the primary, and bent it out of shape. It also broke 2 discs on the secondary. I raised the secondary up about 3”. I tried my new code, and it worked again. I used up to 16 cycles, and hit the wall at 97.5” I finally showed the coil to my new friends on Halloween. At 16 cycles, it was hitting the ground a lot, and the primary on occasion. I also got a couple pri-sec arcs. Of course, I got greedy, and made the max 20 cycles. I never got that high though. As I was turning it up, I got a primary strike, and it shut down.
I looked at it today, and the power trace had fused open directly under the secondary. My thinking is that this time, the strike had missed the strike rail, and caused a voltage spike across the mains. That caused it to arc across the short distance on the pcb, and short the mains. Of course, this means open circuit within a very short amount of time. Hopefully, the uC isn’t damaged.
I plan to construct a better strike rail, and run this secondary to destruction. Then I’m gonna send Mr.Ward some magnet wire, and he’s gonna wind me a new one 6” longer J
The problem was just that the trace under the connecter had been fried. The primary strike caused a voltage spike that shorted the mains. Luckily, nothing else was damaged.
I did construct a better strike rail, and got a good picture of it working. I added 3 extra rails; one near the top, one at the bottom, and one about 2/3rds up.
When it struck to the strike rails, or ground, it made a funny noise. It missed bangs or something. I started at 77”, and got there at 9 or 10 cycles. I moved it out to 97”, and got there no problem. It was hitting the ground a lot, but I had more power.
I moved it out to 104”, and when I was turning it up, it hit the ground, and I heard that strange noise for several bangs, and then the dreadful ‘ping’ sound that tells me I lost an IGBT.
It turns out that there was a ‘reset’ pin on the µC that my dad never told me about. It has a trace going to the connector for programming, but no pull up resistor or anything. It was open. I put a 22nf decoupling capacitor, and a 150k pull up resistor on it. This fixed that problem. I turned it up, and got some good ground and primary strikes. I found that it tuned best at the end of the primary. If I tuned it slightly low, the sparks disappeared completely. The current would build up, and at 14 cycles or so, start dropping. The most current it saw put 12.4 v peak to peak on the scope. The current transformer got messed up somehow, and is only putting out a few hundred millivolts now. The sparks seemed like they started getting shorter, and my mom noticed it too. The caps were hot, and smelled like burning (bad). The capacitance of the cap banks were too high; 1.73µF and 1.81µF instead of 1.2µF. not tonight, but the last time (nov 7?) I measured the temp of the MMC, and got 125 degrees Fahrenheit.