GDT keeps on killing IGBTs

[ritaismyconscience]

Tried it at half voltage so far, got a lot of buzzing but no arcs.

I think I'll have to tune it. Frequency is off by around 20kHz

I used about 900-1000 turns of 26AWG on my secondary, which seems a bit low. Will this matter?

[Zipdox]

I saw this vid yesterday and though of it now

[ritaismyconscience]

Well, it looks like it's making sparks but very small.

Definitely needs more tuning.

Tests were done at half voltage (170V)

Edit: Made primary 3.75 turns instead. Now, frequency is about 190kHz. Think the extra wiring conencting the primary to the bridge is causing a bit extra inductance. Will try to reduce tomorrow.

Here's a test on pulse skip mode, which seems to make longer arcs

[davekni]

Yes, with a coupling factor on the low side, accurate tuning is more critical.

[ritaismyconscience]

I'll try moving the secondary down a bit too.

Could anyone give me a few tips on tuning the system? I'm going to get my oscilloscope probes sometime soon.

I think I did the same thing as you (davekni), like I'm using a spiral primary made out of litz wire too. Could you tell me how you tuned yours?

[davekni]

First measure the secondary resonance frequency.  Then run your primary the way you had been without the secondary, and adjust its frequency slightly below the secondary frequency.  How much below depends on coupling factor and how long arcs will be (how much the arc will add to top-load capacitance and lower secondary frequency).  I'd start with the primary frequency only slightly below the secondary frequency.

The easiest way I know to measure secondary frequency is with a signal generator.  Do you have one?  Wire it through a resistor to a scope input, say 1k to start.   Connect the scope ground to line ground and/or your counterpoise or whatever you normally connect the bottom of the secondary to.  Connect the secondary bottom lead to the scope probe and 1k resistor.  Make sure the top load is in place and connected with whatever breakout point you are using.  Then adjust the signal-generator frequency until the voltage drops.  The frequency where the voltage drops the lowest is the secondary resonant frequency.  For increased accuracy, after you've found the frequency roughly, you can increase the 1k resistor to 10k.  That will make a sharper amplitude dip at the resonant frequency.

[ritaismyconscience]

My signal generator isn't really good (it's one of those cheap analog ones)

I'll probably get one of those AD9850 boards to get a better frequency generator.

[davekni]

As long is you can adjust frequency reasonably smoothly, it doesn't matter what the quality of the waveform is.  Sine is best, but square or triangle can work.  The amplitude notching will look a bit strange, as only the fundamental harmonic is notched, while the higher harmonics will remain.  It's still easy to find the notch minimum in spite of the remaining harmonics.

An alternative secondary frequency measurement technique is ring-down.  Place a ~1nF capacitor across a scope channel input.  Then take the secondary ground lead, touch it to a somewhat high voltage (ie. 170V) source, then touch it to the scope input.  The scope should see a ring-down signal of a few volts.  That ring-down will be at the secondary resonant frequency.  You will need to trigger on the ring-down signal itself, and have the scope persistence set long enough to view a single-shot trace.

[ritaismyconscience]

Using your method, I got a resonance of 171kHz when measuring without a wire attached. With a roughly 5 foot wire attached, it dropped to 150kHz.

Here's another method I tried:

(skip to 5:58)

I got 224kHz and 208kHz using this method. Is there something wrong with this one. The guy in the video basically uses the secondary and his scope probe as antenna, and he measures the frequency where the scope picks up the most signal.

[davekni]

That video method is very similar to the first method I suggested using a signal generator.  The only difference is that the video uses the peak of the top-load field rather than the notch in the drive voltage.  Any of these methods should work fine.  The signal-generator methods (both mine and the video's) are simpler to run.

The difference you are seeing between signal-generator and ring-down is to large to be just random measurement error.  One point comes to mind:  The video makes an implicit assumption that both the scope and signal generator are grounded to line-ground.  If either one is floating, the frequency will read higher than in actual operation.  To be certain, you could explicitly wire the scope ground lead to the signal-generator ground (to the other side of the generator output that the video says to leave floating) and then the two together to line ground.  Depending on how close other grounds are to your top load, connecting the primary winding to ground may make a difference too.  Grounding the primary coil will further improve accuracy, as it will be close to ground potential compared to the top-load's high voltage.

[ritaismyconscience]

You're right, tried it again with the signal generator grounded and got 171kHz and 150kHz.

Looks like I need 5 and a third turns. Will try tomorrow.

I tuned the primary to 156kHz. When I try pulse skip mode at 25% of the max voltage, I get a pretty big spark. When I try the other mode, which runs it normally with small pulse width, I get much smaller arcs, even at 50%. What pulse width should I use for the interrupter?

Also, do I need to increase the coupling?

[ritaismyconscience]

Looks like I'm pretty close to being done

Here's a test at roughly 50-60% max input voltage. I can't go higher because don't have enough space. I'm running this on pulse skip mode, and I lowered the secondary to around 1 inch below the primary to increase coupling.

Edit: Test at 75%

Looks like I'm not getting big arcs because it's arcing to the bushes in the back.

[davekni]

Nice to see!  All the frustrations of the process fade when things work in the end.

[ritaismyconscience]

I ended up blowing four of the IGBTs when I ran it at 90% input voltage. An arc formed between the secondary and the ground rail, and that seems to have caused the IGBT to blow.

Should I lower the overcurrent? I have it set at 500A right now. I'm using 8 IGBTs in total, so it should work (each of them is rated for 225A pk in the datasheet)

I think I need to lower the strike rail too.

[davekni]

That's disappointing.  Hopefully the fried IGBTs didn't take out anything else.

Was the arc from the top-load to the ground rail, or from some point part way up the secondary?  The latter is the issue I had when my secondary started 50mm below the primary.  Raised the secondary to only 25mm below and added more polyurethane coats to the secondary.  Yes, lowering the grounded strike rail can help, but keep it slightly above the primary coil.

Two possibilities come to mind.  First is that the high-frequency noise generated by a ground strike coupled into circuitry, perhaps confusing the driver, making gate voltage transitions away from the current zero-crossing points.  Besides generally keeping wires short etc., make sure the ground connection wire for the strike rail doesn't pass close to drive circuitry.

The other is what I mentioned way back in reply 25 on March 1st:  Last I saw, you have relatively long and separated wires from your bulk caps to the H-Bridge.  That inductance can cause problems at the end of an enable pulse, when the H-Bridge changes from pulling current from the bulk caps to pushing it back as it removes energy from the primary tank circuit.  That issue forced me to redo my bulk-to-bridge wiring to reduce inductance.

[ritaismyconscience]

Could it have been overcurrent? I think the current rises quite a bit when the secondary shorts. I've heard the OCP might be slow and take a few cycles to actually trigger. In that case 500A might be too high.

I'm also using pulse skip mode, so the driver tries to keep the current at 500A. Maybe the current increased too much when it shorted.

The arc went from topload to strike rail. At 75% I was noticing corona discharge from ground strike rail.

For your comments, I don't think the wire connecting the strike rail to ground passes close to the driver. The driver is also in a grounded cage. I'm using the mains ground, will that cause problems?

Putting the capacitors closer might be difficult because I'm using a voltage doubler and I have to try to share the current equally between the capacitors. Maybe I could add 1000 or 2000uF more capacitance close to the IGBTs to solve this issue.

Also, for my next project, I might use the 120N60 IGBTs, which seem to have 600A peak current in a TO264 package. Would this work? Here's the datasheet: https://www.mouser.com/datasheet/2/240/Littelfuse_Discrete_IGBTs_PT_IXG_120N60B3_Datashee-1591813.pdf

[davekni]

Overcurrent is a possibility.  Some designs check only positive half-cycles, so it might be the end of the second half-cycle above 500A before shutdown.  Pulse-skipping could make that worse, as you're then hitting that current limit every few cycles, rather than once per enable pulse.

Pulse-skipping may be risky for other reasons too.  Depending on how it is implemented, it can make IGBT power sharing uneaven.  It also will make lower-frequency components to the gate-drive signal, which may require higher inductance and volt-second rating of the gate-drive transformer.  Or, if it puts some gates into the off 0V state, then there's extra half-voltage gate transitions, which are slower turn-off and may have some momentary weak turn-on overshoot pulses.

In short, I'd disable pulse-skipping mode until you have a chance to examine it closely with your scope at lower bus voltage and lower current limit settings.

The mains ground is probably fine.  I prefer to use both the mains ground and an aluminum foil (or screen or plate) ground plane under the coil against the floor or ground.

Additional 1000uF near the H-Bridge would be good.  Or, use lower inductance wiring from your existing bulk capacitors.  Multiple twisted pairs is good - that's what I needed to do even for a fairly short run.  Each pair has one wire VBus+ and one wire VBus-.  Even if you go with an additional 1000uF closer, I'd recommend twisted pairs between it and the H-Bridge.

The IGBT part you listed doesn't have an internal anti-parallel diode, so isn't a good choice for H-Bridge use.

[ritaismyconscience]

The left side of the half bridge didn't suffer any damage, but 3/4 IGBTs on the right side blew. I replaced all 4 just to be safe. None of the other pars seem to have failed, when I replaced the IGBTs, everything worked again up to 75% input voltage.

I lowered the strike rail, so I haven't gotten any strikes yet. I shorted the secondary and ran it at 50% input voltage, and I used your method to look at the voltage at the capacitor. It looked like a sine wave but it was considerably more fuzzy.

Also, during a ground strike, wouldn't the inductance of the primary drop, making the resonant frequency higher? Could that be a problem (like could it cause the IGBTs to switch at the wrong times?) I think the amount of phase lead produced by my circuit varies with frequency, and my phase lead circuit seems to work better at lower frequencies.

The IGBTs I'm currently using also don't have antiparallel diodes, so I'm currently using 4 fast diodes in place of them. Could this be a problem?

[davekni]

Are your fast diodes on heat sinks along with the IGBTs?  If not, are you sure the diodes are all still good?  I'm wondering if the diodes got hot enough to either fry or have extreme leakage current and slow switching to the point that it fried IGBTs without the diodes themselves failing.  Skip-pulse mode puts more heat into the diodes, as does secondary arcs.

Yes, DRSSTCs usually run at the lower of the dual resonance peaks of the coupled resonators, so a shorted secondary will raise the frequency.  The frequency change generally isn't significant enough to cause problems, but might be if your phase-shift circuit is particularly sensitive to frequency.  Usually the more significant effect is the increased Q of the primary, since the secondary no longer uses much power.  That causes the primary current to ramp up relatively quickly and ramp down slowly.  The pulse-skip mode will then need to skip lots of pulses, moving more power to the diodes, and likely causing more issues with gate-drive low-frequency components.  Even without pulse-skip, there will be more power for the diodes during the decay after the enable pulse ends.  (I'm working on a circuit to detect ground arcs and terminate the enable pulse before current ramps up too much.  My IGBTs have internal diodes, but their power dissipation capability is much lower than the IGBTs themselves.)

The other possible issue with external diodes is inductive voltage drop between the IGBTs and diodes.  I don't have any personal experience with external IGBT diodes.  I think IGBTs are usually rated for ~20V reverse Vce.  I don't know how sensitive they may be to spikes beyond 20V.

The "fuzzy" sine wave was likely multiple captures as the primary current ramps up.  The key reason to look there is for accurate phase adjustment.  Typically you can see the sine wave along with small steps at the top and bottom when the H-Bridge switches.  Those steps should be just barely before the top and bottom crests of the sine wave.  Try triggering on the enable pulse to get a stable waveform, or set the scope's trigger hold-off longer to prevent multiple triggers per enable pulse.

[ritaismyconscience]

I successfully got my coil to work at full voltage without blowing up:

I checked and the overcurrent LED never lit up. I'll have to measure the pulse width, but it's set to 75% of the max pulse width the controller lets me set.

Edit: tried it at the highest pulse width, still no overcurrent. I think my transistors blew up because of overcurrent. Could also be faulty OCP?

Actually, I think I need a higher capacitance. Right now I'm using 100nF, I should probably use 200nF or something.

150nF would get me 4.5 turns and reduce the surge impedance to 7 ohms (from 10)

200nF would get me 3.75 turns, surge impedance = 5 ohms