GDT keeps on killing IGBTs

[ritaismyconscience]

Looks like I found the answer

https://www.microsemi.com/document-portal/doc_download/14696-igbt-tutorial

Pages 2-3 say that high collector current could cause the IGBT to latch on.

I'm wondering if I should go back to using 100nf for primary capacitor since the primary current rises really fast

[ritaismyconscience]

Got it to work at full voltage without blowing up. Set overcurrent to 500A.

I'm only getting 1 foot arcs, which means that my coil is probably a bit out of tune.

[davekni]

Yes, for 500A and ~150kHz, 100nF MMC is probably better.  Hopefully that will increase your arc length.

[ritaismyconscience]

Looks like you're right.

Tested at half voltage, already working really well. Will have to go somewhere bigger to test.

Edit: This one's probably better. Moved the breakout point up a bit so it doesn't get cut off by the door.

Thanks for your help davekni!

[davekni]

Great to see success!  When you can test in a larger space, you'll find that higher pulse repetition rate extends the streamer length well beyond that of single pulses.  Hopefully everything is robust enough to handle higher rates for an impressive show.

You are certainly welcome.  I'm glad to have been of help.

[ritaismyconscience]

Just out of curiosity, is there a reason why 100nF works better than 200nF?

[davekni]

Yes.  For your fixed frequency and fixed 500A current limit, 100nF has twice the impedance, so twice the primary voltage and twice the primary energy.  It also takes longer to ramp up to the 500A limit.  The short ramp at 200nF may not be long enough for the primary energy to transfer to the secondary, besides having less primary energy available to transfer.

Are you still hitting your 500A limit on most pulses?  If so, even lower than 100nF might be better.  Of course, going lower requires your primary to handle higher voltage.  An arc across the primary coil will destroy IGBTs when the MMC suddenly discharges into the H-bridge.  Also be careful of running longer pulses, as the IGBTs may not handle just under 500A for the longer times.

[ritaismyconscience]

I'm already tripping the 500A limit at around half voltage. I'll try reducing the tank capacitor to 50nF.

[davekni]

50nF sounds good, but do be careful to have sufficient voltage capability.  At 150kHz, 50nF has 21.2 ohms impedance, so will have 10.6kV at 500A.  Plan for at least 12kV to have margin for any delay in over-current shutdown.

[ritaismyconscience]

I changed it to 50nF and it seems to work better. Haven't tested it at half voltage yet.

Some pictures:







I guess one of the things I could change is moving the FB and CT coils onto the lower voltage side of the MMC. However that means my leads have to be longer.

[davekni]

Yes, the current transformers should definitely be on one of the H-Bridge outputs, not on the MMC-to-primary connection!  Don't want to rely on CT insulation holding up to 10-12kV!  CT lead length isn't as critical as for gate-drive, but I'd  still suggest seeing if you can rearrange your layout to keep leads short.

[ritaismyconscience]

It seems like there's a few problems with my phase lead circuit. First of all, at low currents(~40A), the duty cycle is a few percent less than 50 because one of the diodes (the one connected to 0V) clamps more often than the other. Also, my circuit seems to also be current dependent (higher current results in smaller phase lead). I'll probably have to adjust it while testing at higher current.

I can't use the other circuit for phase lead because the rising and falling edges are not really steep, and the extra time it takes to reach the HIGH threshold basically offsets the phase lead.

edit: tried measuring, ended up having a spark form between primary and oscilloscope probe. Nothing broke (pretty sure there's a MOV inside the oscilloscope that saved it)

[Mads Barnkob]

Yes, the current transformers should definitely be on one of the H-Bridge outputs, not on the MMC-to-primary connection!  Don't want to rely on CT insulation holding up to 10-12kV!  CT lead length isn't as critical as for gate-drive, but I'd  still suggest seeing if you can rearrange your layout to keep leads short.

The current transformers (FB/OCD) should be on the primary lead from inverter to primary coil. As Dave mentions, between primary coil and MMC there is a very high voltage present due to resonant LC voltage ring up. But even on the inverter to MMC connection you have some higher voltages than on the other leg.

It seems like there's a few problems with my phase lead circuit. First of all, at low currents(~40A), the duty cycle is a few percent less than 50 because one of the diodes (the one connected to 0V) clamps more often than the other. Also, my circuit seems to also be current dependent (higher current results in smaller phase lead). I'll probably have to adjust it while testing at higher current.

I can't use the other circuit for phase lead because the rising and falling edges are not really steep, and the extra time it takes to reach the HIGH threshold basically offsets the phase lead.

edit: tried measuring, ended up having a spark form between primary and oscilloscope probe. Nothing broke (pretty sure there's a MOV inside the oscilloscope that saved it)

40A or low test voltages are not sufficient for testing a Tesla coil. You need to have the system in a state of consistent energy transfer where different components minimum values for "linear" behavior is met. Like low voltages across a IGBT gives you a much higher output impedance or feedback signal is too weak for the driver to properly lock on in the heavy EMI.

Some 100-200 VDC bus and at least 100-200A in the primary circuit to even begin to think about looking at phase lead, you just risk adjusting the phase lead for the low voltage test conditions that are far from what you see at full power. Often you have your primary circuit detuned lower for longest sparks, as the secondary circuit goes down in frequency as the sparks grow longer (presenting a capacitive load).

You need to use a differential probe when measuring on inverter outputs to avoid damaging your oscilloscope. You risk high voltage transients or grounding the power circuit through your probe.

[davekni]

Mads,
When you said "But even on the inverter to MMC connection you have some higher voltages than on the other leg.", are you referring to the other H-Bridge leg?  Hopefully the two H-Bridge outputs have the same voltage (180 degrees out-of-phase) or some electronics has fried.  The only difference that comes to mind is parasitic wiring inductance and the MMC's parasitic capacitance to ground, which wouldn't matter if the CT is close to the H-Bridge.

Separately, ritaismyconscience, have you measured your scope probe input impedance?  I had a probe fry in a way that it still functioned, except that it had about 600 ohms from tip to ground.  Took a while and some confusion on my part to figure out the probe issue.  Was initially using it probing a low-impedance signal source, so didn't notice any problem.  Moved it to a higher-impedance node, and results got confusing for a while.

[ritaismyconscience]

Checked and no change in impedance. Think I got lucky becuase I accidentally left it on 50 ohms when it hit. Anyways my scope should have input protection (it's fairly old though).

I measured the voltage by insulating a section with tape and putting a wire next to it (basically I made a capacitive dropper). It ended up arcing over an inch around the tape and onto the probe, but nothing bad happened. Also, I'm having a bit of trouble determining the phase lead with the current setup because the current ramps up in a few cycles and just stays there, and during the part where the current is constant, I can't really tell when the IGBTs are switching and when they aren't. It just looks like a normal sine wave.

[davekni]

That inch arc explains why current-sense transformers should be on one of the H-Bridge outputs and not on the MMC-primary junction.

For phasing, if one channel is measuring current, you need another measuring H-Bridge output voltage.  Using a single probe works when measuring the MMC-primary junction (capacitively since you don't have a HV probe).  Perhaps you should put a small glass jar (or test tube or ...) over the end of the probe for better insulation.  That capacitive-coupling probe method is for high-impedance scope input.  With 50 ohm scope input, you'd get the derivative of the voltage (high-pass-filter).

[ritaismyconscience]

Think I got phase lead set correctly



Here's my setup. I probed the H-bridge to Primary wire.

[davekni]

Is that a current or voltage waveform?  If current, the only slight glitch that might indicate switching time is at peak current rather than zero.  If a voltage waveform, the H-Bridge output should be a square-wave, not a sine wave.  The sine-wave voltage I'd suggested probing for phasing is at the MMC-to-primary node, probed through good insulation (ie. small glass jar).

[ritaismyconscience]

I'm measuring voltage, not current. I think I probed the two H bridge outputs with the method you described, and one of them looks like the sum of a square and sine wave, and the other just looks like a sine wave.

[davekni]

Both H-Bridge outputs should be square waves, except for the vary tail end of ring-down after a burst ends.  The sine + square is the junction between MMC and primary coil.  That's where the voltage is high and you can get the 1" arc, so why good insulation is necessary.