DRSSTC Overvoltage Issues

[Max]

I planned on making a thread about my DRSSTC for a long time but I always postponed it because I still had to fix this and change that. Now, after it exploded on me for the third time I finally write this post. While it's still kind of a presentation it's mainly because I have not the slightest idea of what's going on and hope you guys can help me. But be warned, to give you a complete picture, I got to go back in time a bit because I've had issues with this setup since first light - which was 4 years ago.
To keep the length of this post somewhat reasonable I only link to the pictures and videos instead of including them within the post. While I never talked about the build process here, I documented everything in the German equivalent of this forum (mosfetkiller forum).

Just to give you an idea even if you're only skim reading, here's one pic/video (the one of the recent test and failure):

Current specs

• When designing the coil I aimed for 32A 400V 3 phase operation.
• Filtering: 1,5mH frequency drive sine filter on each phase. Boosts the idle voltage from 565V to ~650V.
• Rectifier: 3x 35A 1kVdc fullwave modules, only using "half" of each one.
• Full bridge made of 4x SKM200GB128 300A 1200V half bridge modules paralleled in pairs
• Laminated bus bar design. Multiple layers of 2mm aluminium and silicone sheets. held together by Nylon M6 screws.
• 10x 3300uF 450V Epcos bus capacitors (measuring 3000uF and 20mOhm ESR). 5 parallel "strings" of 2 in series for ~8000uF @900V and 250Arms
• 4x 680nF snubber capacitors distributed across the bridge
• MMC: 14x 5uF 1200V Arcotronics GTO snubber for 360nF @12kV, 1.6kApk, ~160Arms
• Primary: ~5 turns, made from 10mm copper pipe
• 10k resistor across primary tank
• Secondary: 20cm x 105cm, ~1700 turns, 0,6mm
• Topload: 9x 25cm balls arranged as ~1m "toroid"
• GDT: ~60x40x20mm (outer dia. x inner dia. x height) N30. 11:11:11:11:11 winding/ratio. 1 GDT per pair of halfbridge.
• Gates: About 13nF gate capacitance, 5R1+1N5819 antiparallel diode, 24V bidirectional TVS diode. Gate signals show ~600ns/400ns rise/fall time. Only one overshoot clipped to 24V, stable plateau at 22V
• Driver: UD+, set for 900A pulse skipping.

The issue
After the most recent events it seems to me like all the issues I've had over the years could be explained by one root cause: serious overvoltage on the bus. And with serious I mean "visible 5mm blue arcing" overvoltage. However I don't have the slightest idea where that would come from.

History of events
When I started with this project 5 years ago my only power electronics experience was an SSTC which I never really finished and never worked realiably. I had read a lot before starting but tbh I didn't know half as much as I do today.

About two years ago I made a partial redesign of the setup to fix a couple issues I had encountered til then. Here are the important differences compared to the current setup:

• GDTs made from CAT5 patch cable as I had seen this being recommended at various places. Of each twisted wire pair I took the white cable and connected all of them together to form the primary. Every colored cable was connected to one gate.
• No input filter or choke
• Different bus bars with laser cut polycarbonate isolation. Creepage distance of 2mm. An early CAD model can be seen here.
• The bus capacitors were all paralleled, meaning not 5 strings of 2 but 2 groups of 5 parallel.
  
• 2x 440V TVS diodes in series across each IGBT pair (mounted inbetween the modules, visible in the pic linked at point 2 below)
• MMC made of 15 (instead of 14) GTO snubbers for only 330nF instead of 360nF
• Thinner wiring in the primary circuit

Pictures of the old setup

And finally, the list of important events/failures over the years:

• Popping and crackling noise from the bridge. Solved by cleaning off the burns from the laser cut isolation between the bus bars that likely caused small arc overs.
• Explosion of an IGBT module when first trying 3 phase (400V) operation (static load, very low power). Exact cause undetermined. I found signs of arc over between the power rails.
• Redesign. I spent a lot of time designing this in CAD and thinking about how it would be assembled. Turned out I was wrong and the screws cannot be reached. Don't ask me how I assembled it (and disassembled when it went boom). Let's just say it took hours.
• Burnt GDT. When taking out the GDT I found that it had burned through the isolation where it touched the grounded heatsink. Dismantling it I found another spot where the isolation of one primary-secondary twisted wire pair had molten away - only <1mm of air was left as isolation. No clue why this happened. Rewound it with individual wires and potted afterwards.
• Power run on 3 phases with the new setup and a static load, measuring temperatures and other things. Went fine for about 15 minutes. Then, when refiring it after an hour of cooling down, an arc over between positive and negative rails occured. Damage: One highside and one lowside IGBT in one halfbridge pair failed, one of the modules had a crack in the case. Both modules got replaced. Steve Ward thought it was likely that one IGBT failed first and the resulting flash would have shortened the rails. He also was the first one to bring up the insufficient camping force of the Nylon screws. For some reason neither me nor anyone else previously thought about that. I somewhat doubted that the IGBT failed first. Why would it? It didn't fail during the high power run. I saw once again spotweld marks on the rails and came up with the suspicion that the Nylon screws - especially after having heated up and cooled down - were insufficient for a good contact between bus bars and IGBT terminals, causing a spark which in turns caused the arc over between the power rails. As a temporary solution I added another piece of isolation such that no spark from one terminal could reach another one.
• Another run of the coil (with secondary) on 400V went fine. Gave my best to tune it but the output was not great. This is another issue; since the redesign I haven't had the same output as before. Previously I got bright >2m arcs with 230V 16A single phase input. Now I needed 400V three phase for a similar output (though 3 phase is notably louder than single phase - even with less output).
• First test run (with secondary) with my Syntherrupter. All events cited above had been done with its precursor (no polyphony, no special effects, so not much to go wrong). The test run went rather well until the coil once again went boom. This time - to my surprise - the arc over occured between one of the rectifier terminals and its case. I'm pretty sure that it's not the rectifier itself that failed first because the potting shows absolutely no signs of cracking as you'd expect it if >1kJ would go though it. The terminals and the case however show clear burn marks. Nonetheless, the rectifier was dead. I couldn't explain myself this event. The previous theory can't explain it and I couldn't see how an arc over at this place could occur without any IGBT breaking down.
• I rewired the rectifiers (see quick specs, I have enough redundancy for such a failure) and slowly restarted testing (on 230V single phase though). The output was not consistent; I can only describe it as "stuttering" (I can upload a video of it if someone's interested). Increasing the power did not really help. I searched for other issues but couldn't spot anything. To get an idea I hooked up my scope to the bridge outputs (1 chn per output). Without the ability to scope the interrupter signal or primary current (didn't have the necessary parts on hand) The scope shot was pretty useless as I had nothing to trigger on. However, the scope did make another big difference: the issue was now gone. It even seemed to me as if I'd be getting more output now than before the bang (more similar to what it used to be before the redesign. I have no explanation for this but after a bit more testing I ended it there.
• Few days later I was supposed to do a demo of the coil. However, because of bad weather and other reasons it should be a static load only (using it as improvised induction stove; we needed one). Again I heard that same stuttering, but this time - without the loud arcs - I could hear that it actually sounded like arcing around the bus bars. And after looking around for a while I found that indeed I could see it arcing between positive and negative bus bars. Small blue arcs that matched the fizzling/stuttering sound. Since I reacted immediately when I saw it I failed to take a picture or video of it and can't tell you whether it was inside or outside or through the silicone isolation. I was absolutely speachless. Why the heck are there even such voltages?? How comes that it doesn't cause a short? How do the IGBTs survive that? With easily 5mm long arcs all the previous event's aren't surprising anymore (though I never noted such fizzling/stuttering except for the very first point of this list).

And here we are. Sure enough the Nylon screws are a known issue and tbh I was expecting another failure related to them but didn't care for the time being. I do have a new design ready to manufacture (with steel screws and easy assembly) but til I'd find the time to actually build it I didn't mind risking another crash (maybe because I have >20 replacement IGBTs...).
Anyways, I cannot see how high contact resistance would cause such overvoltages. Neither can I think of any other reason for that. Why does a scope with 10MOhm probes affect any of all this? Also, why did it seemingly increase the output? These recent experiences make me feel like I'm missing some important knowledge when it comes to DRSSTCs... So I hope you guys got some ideas.


Thanks for reading!
Max

[AstRii]

It could be that one IGBT fails open circuit while there is a lot of current flowing through primary, this insane dI/dt can easily create high voltage to spark over a long distance.

Have you tried decreasing coupling? There could be some flashovers between primary and secondary. If quick enough, they would not be even seen in camera, especially in bright room conditions.

I would also get rid of those 880V TVS diodes. They look as a protection but they really do more damage than healing. Think about what they do, they basically short your bridge each time you have an overvoltage event.
Some 24V TVS across G-E to protect gates are fine, since your GDT will not provide much current to overheat the diodes so quickly. But your DC bus caps? They will push 10s of kA through them which can destroy them in an instant and short your bridge afterwards.
I have destroyed a DRSSTC bridge this way already, using 440V TVS on 325V bus.

[davekni]

I would also get rid of those 880V TVS diodes.

Agree.  No personal experience, but have seen multiple cases of TVS-induced failures here.

I suspect the real issue is aluminum.  Replace all aluminum interconnect with copper.  Aluminum is used for power wiring, but with anti-corrosion coating on connection surfaces and high clamping force.  Aluminum's native surface oxide layer is extremely problematic.

900A is enough for noticeable conductor vibration due to magnetic field.  Makes interconnect electro-mechanical reliability more important.

Remote chance you could have bad luck with bus bar inductance and snubber capacitance resonating at 2x coil operating frequency.  I've seen that once here.  Bus ripple current is at 2x coil frequency.

[Max]

Thank you for your input!

It could be that one IGBT fails open circuit while there is a lot of current flowing through primary, this insane dI/dt can easily create high voltage to spark over a long distance.

That might explain a single failure but not sustained arcing?

Have you tried decreasing coupling? There could be some flashovers between primary and secondary. If quick enough, they would not be even seen in camera, especially in bright room conditions.

Hard to do with a static load. Most failures occured without secondary.
But no, I never tried. Will try on the next test run; curious to see if and how it changes the output.

I would also get rid of those 880V TVS diodes. They look as a protection but they really do more damage than healing. Think about what they do, they basically short your bridge each time you have an overvoltage event.

First of all, they got removed with the redesign. But to be honest, I don‘t quite understand the reasoning here. What you say makes sense to me for devices like a gas discharge tubes which - once it conducts - stays contuctive down to very low voltages. TVS and MOVs on the other hand I thought wouldn‘t do that. And since the bus capacitors aren‘t charged to a voltage much below the breakdown voltage they shouldn‘t contribute any current to the TVS diodes. The only risk I can see is if they fail short circuit. Then they‘ll be vaporised for sure.
Btw I never had a failure of a TVS diode before they got removed.

I suspect the real issue is aluminum.  Replace all aluminum interconnect with copper.  Aluminum is used for power wiring, but with anti-corrosion coating on connection surfaces and high clamping force.  Aluminum's native surface oxide layer is extremely problematic.

I guess with decent clamping force it wouldn‘t be that problematic. The only signs of such issues appeared at the IGBT terminals. All other aluminium connections were fine. At least that was the case for the previous version which I dismantled a few times. I haven‘t dismantled the bridge after the latest failure yet.
Anyways, since I got some scrap copper recently I do plan on making the next version from copper.

900A is enough for noticeable conductor vibration due to magnetic field.  Makes interconnect electro-mechanical reliability more important.

A quick calculation suggests forces in the order of magnitude of 40N/m or about 15N for my bus bars. Not sure that‘s sufficient but I‘m willing to believe it; it could kinda explain the recent observations. Though I‘m still surprised that voltages high enough for arcing don‘t kill any semiconductors…


Regards,
Max

[Intra]

Pictures of the old setup

Hi, Max! You have your driver not in full cover aluminium box on this photo. Try grounded full cover box.

Also it seemed your UD+ is on breadboard? Please post photos of driver part from opto to igbt.

[Uspring]

Filtering: 1,5mH frequency drive sine filter on each phase. Boosts the idle voltage from 565V to ~650V.

What does your frequency sine wave filter look like? Possibly it provides a bad RF ground to the bridge. Capacitive coupling of the high voltage on the primary tank to ground might lift the bridge and the bus up to a high potential. That would be a common mode voltage on the bus bars and not a differential one, though.

[davekni]

First of all, they got removed with the redesign. But to be honest, I don‘t quite understand the reasoning here.

The issue is that TVS diodes usually fail-to-shorted when over-stressed.  Typical IGBT die are much larger and can handle much higher avalanche energy than these relatively-small TVS diodes.  A diode fails first, then the opposite IGBT fails driving into the short.  It is a bit like the old joke about transistors sacrificing themselves to save the fuse.  May not be a factor in your specific failures.  Clearly not now with them gone.

Intermittent aluminum connections may be triggering failures, creating voltage transients and/or tiny arcs that ionize enough air to start a cross-supply arc.  (Forgot to mention earlier: Most aluminum wire connections use knurled or otherwise textured surfaces to increase pressure for given total force.  This causes small localized deformation of aluminum to break through oxide.  Then maintained force and anti-corrosion compound minimize new oxide growth.)  Great to hear that your new version will be copper!  At least, if there are any remaining issues, they will be easier to find with consistent connection quality.

A quick calculation suggests forces in the order of magnitude of 40N/m or about 15N for my bus bars. Not sure that‘s sufficient but I‘m willing to believe it; it could kinda explain the recent observations.

Yes, not a large force.  My concern was resulting mechanical vibration at interrupter frequency loosening connections.  Suppose there is a remote chance of vibration frequency hitting some mechanical bus-bar resonant frequency, resulting in larger deflections.

Burnt GDT. When taking out the GDT I found that it had burned through the isolation where it touched the grounded heatsink. Dismantling it I found another spot where the isolation of one primary-secondary twisted wire pair had molten away - only <1mm of air was left as isolation. No clue why this happened. Rewound it with individual wires and potted afterwards.

Did you recheck gate waveforms after GDT reconstruction?  Depending on relative winding placement, leakage inductance can be much higher with non-paired primary/secondary wires.

[Max]

@Intra: good catch! That‘s actually something that changed already before the first crash. I forgot about that. All crashes occured with an UD+ inside a metal box.
That breadboard thing was a UD2.7 btw. Design pics/files available here.

@Uspring: sorry, should‘ve been more clear. It‘s just three big, individual inductors. One per phase. They‘re visible here.
While grounding is something I‘ve been struggling with, too, it can‘t be the origin of the arcing since that was with a static load.

@davekni: Thanks for the explanation. I did know that short-circuit failures are problematic for the reasons you gave. I just didn‘t know they fail that often.
Mechanic resonance is unlikely; failures - including the recent arcing - occured at different frequencies.

I could of course just rebuild the whole thing for the third time and hope it magically fixes the issue. That’s kind what I did the last time. Identify the most likely cause, fix that (among other things) - and find out that the issue is still there. I don’t want to rebuild it a fourth time…


Kind regards,
Max

[Intra]

@Intra: good catch! That‘s actually something that changed already before the first crash. I forgot about that. All crashes occured with an UD+ inside a metal box.

Then it will be good if you show fresh design.

Also, for 75kHz fres, Ae=200mm^2, 240mT of N30, and 24v (if it so), it good to use 4 GDT windings but you have 13. What's on scope?

Also, pvc screw can get warm from aluminium bus and loose conduction from bus to igbt\cap. I recommend brass screw.

And rigid FR4.

[Max]

@Intra: here it is. The driver sits in the box below the PSU. Right now the DRSSTC sits in the storage room so I can‘t take it apart to show you the UD+ inside the box. For the same reason the pictures aren‘t of great quality; not much light there. I have to say tough that I can‘t remember to have had issues with the driver itself. The inrush current limiter (attiny + mosfet + relay + resistor) used to have EMI issues (attiny resetting) before I had that metal box. But since it‘s sitting in there with the UD+ I can‘t remember to have noticed anything suspicious.

For the same reasons I can‘t take it apart I also can‘t take scope pictures of the gate waveforms. Therefore I gave a quick description in the specs section of the first post. From all I know they look fine.
Thank‘s for the calculation though. I‘ll remember it for the next time to try much lower turn counts!

By now I know about the screws getting warm and not being strong enough. Mentioned it in my first post under point 5 and at the very end of it.
I am willing to believe you that that‘s the cause of all of my issues but I still can‘t see see how it would explain them - especially the recent ones.


Kind regards,
Max

[Intra]

Sparks like that can happen by many reason. Several most likely is:
Your busses, polycarbonade or air between them for some reason become not enough dielectric to the task after start.
You got gate driving issue and halfbridge got opened to discharge a bus cap to brick. it should be seen on scope.
It happened because of paralleling igbt by lag about parasite inductive between neighbour bricks.

As 300A bricks can handle about 5kA peak current, why use parallel bricks with 20x105 secondary?

What about attiny current limiter? Why don't use just current transformer as usual? mcu in hv field is not stable and it serve what exactly purpose?

Without scope there is not much to say.

[Max]

 

Your busses, polycarbonade or air between them for some reason become not enough dielectric to the task after start.

It‘s been silicone for the last two years; polycarbonate was before. Silicone based layout is linked in the specs section. Anyways, neither one should break down that easily at 2mm thickness so I assume it‘s the air. Even then I got 6mm of creepage between these rails and during assembly I thoroughly cleaned all surfaces. Which is why I was so surprised to see these discharges.

You got gate driving issue and halfbridge got opened to discharge a bus cap to brick. it should be seen on scope.

Not quite sure what you mean by „discharge a bus cap to brick“ but I can guarantee you that the bus caps do not discharge any significant amount of energy into those small arc overs. They‘re thin, blue, sporadic. Nothing energetic. Unlike the arc over at the rectifier; that one did shorten the bus caps and you can see the results in the video.

It happened because of paralleling igbt by lag about parasite inductive between neighbour bricks.

Interesting… I paid attention to keep everything as symmetric as possible for the paralleled bricks - again, can be seen in the link in the specs section.

As 300A bricks can handle about 5kA peak current, why use parallel bricks with 20x105 secondary?

Uuuhm… I haven‘t heard of any 300A brick that would handle anything near 5kA reliably. CM300 which are quite a bit beefier than my IGBTs don‘t even handle that much.
Most people run my bricks at 700-800A. I think someone managed to kill them at <<2000A; can‘t remember. They desaturate at 1800A. Anyways, taking the 750A as reference, I wanted to have some headroom, not only for slight imbalances between parallel bricks but also because I run the coil at high duty cycles, leaving less thermal headroom for extreme current (and thus power dissipation) spikes. Since the coil is expected to operate in shows for many years I do not want to push the boundaries too much.
For some time I had the OCD set to 1200A but I backed down to 900A because of pulse skipping. Maybe a bit too conservative but I haven‘t had the chance to really push the coil. So far it looks like as if I‘d be limited by mains supply anyways. Again, needs more high power testing which in turn needs more reliability.
Some - not all - examples I‘ve been speaking of are listed here: https://kaizerpowerelectronics.dk/tesla-coils/drsstc-design-guide/igbts/

What about attiny current limiter? Why don't use just current transformer as usual? mcu in hv field is not stable and it serve what exactly purpose?

I wrote inrush current limiter; making sure the 8000uF bus capacitance don‘t blow all fuses.


Kind regards,
Max

[Intra]

It‘s been silicone for the last two years; polycarbonate was before. Silicone based layout is linked in the specs section. Anyways, neither one should break down that easily at 2mm thickness so I assume it‘s the air. Even then I got 6mm of creepage between these rails and during assembly I thoroughly cleaned all surfaces. Which is why I was so surprised to see these discharges.

silicone are flexible. too much ways it can let down a task. it can become easier to squeeze when heated and it can make busses too close to each other because of force from plastic screws. not much to say without experiments.

Not quite sure what you mean by „discharge a bus cap to brick“ but I can guarantee you that the bus caps do not discharge any significant amount of energy into those small arc overs. They‘re thin, blue, sporadic. Nothing energetic. Unlike the arc over at the rectifier; that one did shorten the bus caps and you can see the results in the video.

I said that about this case https://forum.mosfetkiller.de/viewtopic.php?p=282738#p282738
Discharge a bus cap to brick means if any issue in gate signal and you will get opened an igbt which should not be opened at this time. current from electrolytic capacitor then will go through the two igbt with no significant resistance and warm up igbt core causing it to explode.

About driver which is in box. Do you have ferrite chokes between gate drive and gdt?

Uuuhm… I haven‘t heard of any 300A brick that would handle anything near 5kA reliably.

See "Wed Sep 24 2008, 03:42am" in https://4hv.org/e107_plugins/forum/forum_viewtopic.php?p=2&id=54347

I wrote inrush current limiter; making sure the 8000uF bus capacitance don‘t blow all fuses.

Sorry, my bad.

[davekni]

See "Wed Sep 24 2008, 03:42am" in https://4hv.org/e107_plugins/forum/forum_viewtopic.php?p=2&id=54347

Thank you for linking to Steve's fascinating experiment.  (I ran a couple test-to-destruction experiments, on a TO247 IGBT and a couple 0.33uF caps for MMC use.)

[Intra]

See "Wed Sep 24 2008, 03:42am" in https://4hv.org/e107_plugins/forum/forum_viewtopic.php?p=2&id=54347

Thank you for linking to Steve's fascinating experiment.  (I ran a couple test-to-destruction experiments, on a TO247 IGBT and a couple 0.33uF caps for MMC use.)

You're welcome.