DRSSTC I Fail

[AstRii]

Hello guys,
today i finally tried my first DRSSTC on full mains voltage (after like a year ) and it was working fine for like 2 minutes
After few minutes i heard loud bang and the lenght of sparks suddenly decreased to few cm.
I am using this basic inverter schematics.

After i dissasembled it, i figured out that 2 IGBTs (T2, T4) failed into short-circuit and D10 440V TVS diode literally exploded.
So here is what i think what happened, voltage spikes were higher than 440V which allowed the D10 TVS diode to conduct. When the TVS diode conducts, it can pass a lot of current which overheated the TVS diode and before it exploded, it went into short circuit which also shorts the low-side IGBT T4. And i think that the TVS diode was in short circuit when the high side IGBT T2 turned on, which would explain that the IGBTs are also broken,
the momentary shoot-through current destroyed it.
I think this is the only explanation that explains why the TVS diode broke.
I don't think that the IGBTs broke because of too high C-E voltage, i'm using FGY75N60 IGBTs rated 600V and bus voltage of 340V DC.
I don't think that the IGBTs broke because of too high current either since i had OCD set to only 200A to be safe (at this voltage and 60us ontime, the OCD was active pretty much all the times).
When this happened, the streamer hit grounded copper grid protecting the bridge.

It is true that the grounded copper grid is literally few cm away from the IGBTs, and a streamer this close could even open one IGBT.
But that woudlnt explain the broken TVS diode.

Do you guys have any theories what could have caused this?

Also, i'm currently ordering new IGBTs for the coil. I was thinking if i could find better TO247 package IGBTs for this coil.
I don't want to drive them much higher than their datasheet values. And i also think if i buy 1200V sillicon, i would be safe for any voltage spikes that could occur in this bridge.
So i was looking for some IGBTs and found this: IXYB82N120C3H1   (datasheet here): http://www.farnell.com/datasheets/2173925.pdf?_ga=2.244592199.332396150.1598123582-1311482206.1541981099
 These IGBTs are rated 1200V and 320A peak current, which should make my bridge indestructable in my small coil.
They also seem to handle 200kHz (my switching frequency) pretty easily.
What do you think about those?

I'd appreciate any theory about what happened and any opinion about the IGBTs,
Thank you
- Mark

[davekni]

That's certainly a reasonable failure scenario.  I think it's also possible that D10 exploded open, then T4 experienced over-voltage and failed shorted, which then shorted T2.  Over-voltage is common immediately after OCD trip.  Many controllers check OCD on only once-per-cycle, so that over-voltage can end up repeatedly stressing one part of each half-bridge.  Yet another possibility is that the copper-grid strike confused control electronics, causing H-Bridge switching at high-current parts of the waveform.  That would depend on the path from copper grid to ground and from electronics to ground and any places for inductive coupling.  For now, let's presume over-voltage is the issue.

Over-voltage is almost always due to parasitic inductance within H-Bridge interconnect.  Gate over-voltage and other destructive behaviors can occur if gate-drive circuitry doesn't make a good Kelvin connection to the emitter.  (If there is significant trace or IGBT lead length that's common to gate-drive and emitter current from the collector.)  For Vce over-voltage, parasitic inductance can be separated into two segments.  First is from VBus (VCC in your schematic) IGBT terminals to the film snubber caps (U1 in your schematic).  The second is from film caps to bulk electrolytic caps (C1 in your schematic).  Best low-inductance layout is to use copper planes for VBus+ and VBus- (GND).  (I use one side of my ECB for VBus, with break down the middle, so two half-size rectangles for VBus+ and VBus-.  The other side is similar, but rotated 90 degrees, with two half-planes for OUT1 and OUT2.  Small cutouts from the planes can accommodate gate drive.)

My guess is that you ran into issues with film-to-bulk cap interconnect inductance.  That's where OCD shutdown causes trouble.  The coil current is drawing power from VBus before OCD, then suddenly feeding power back to VBus after OCD.  That sudden change causes a ring of the L/C of parasitic inductance and film capacitance.  I had to rebuild my film-to-bulk capacitance interconnect (to reduce inductance) due to this exact issue, fortunately caught on scope traces before anything fried.

Yes, higher-voltage IGBTs is one way to tackle over-voltage issues.  (My preference is to scope signals to find where issues are and fix specific causes instead.)  The diode speed on your new device is of concern - much slower than the diode in your existing IGBTs.  (Slow diodes also caused me rework - in my LLC converter isolated PFC that feeds my DRSSTC VBus supply.)  If you reliably maintain phase-lead in your controller, then diode speed doesn't matter.  IGBTs always turn off while still conducting current, so the opposite diodes conduct and are soft-commuted as the current switches direction.  But, if you ever get a bit of phase-lag, then the conducting IGBT changes to diode current before turning off, and the opposing IGBT hard-forces the diode off.  This causes a huge current spike, causing a corresponding voltage spike as the diode finishes turning off.  Inductance from IGBTs to film capacitors (and within IGBT and film capacitor leads) is the concern here for this high-frequency spike.

Nice printed frame/enclosure!  Mine is built into milk crates:)  Your primary coil (and bottom of secondary) is much closer to you electronics than I'd be comfortable with.  It may be fine, but it can be hard to figure out where ~1kV/turn primary field may couple into control circuitry.  Hopefully other people will have opinions on whether this is an issue.  Perhaps I'm just paranoid of difficult-to-quantify possible issues.

Hopefully my dense post is understandable, perhaps after the second or third read:)

[AstRii]

Thank you Dave!
Somehow i managed to understand what you wrote on the second read already
I have the feeling that it was not the over voltage on C-E that caused this, but simply the shorted D10, i read the datasheet of the D10 TVS Diode (1.5KE440CA) and it actually starts to conduct at 418V already.
Some of those voltage spikes could be easily enough to destroy the TVS.
But that's just my theory, i can be wrong
I think the inductance of the DC bus capacitors and the film snubber caps is not that bad.

The path from DC bus capacitors to snubber caps is pretty thick copper with a lot of solder on it to increase conductivity. It's almost as if it was a copper bus bar.

I have tried to meassure the voltage spikes at high input voltages but however i meassure there is always strong noise from the IGBTs switching and the meassured waveform is unreadable.
I have tried to meassure with twisted cables and common mode choke and it didn't help. After those attempts i gave up and just hoped that the voltage spikes aren't high enough to do any damage (which i still think is the case).

I fail to read the datasheet as good as you do. I can't see any information about the internal diode speed. I have always naively thought that the internal diode is as fast as the IGBT itself and judged a speed of a transistor just
with values for t_on, t_off, t_rise, t_fall.
Anyway i'm thinking of using bigger topload since i get a lot of ground rail strikes, so i guess my new switching frequency will be about 170kHz.
I'm using UD2.7 as a driver for this coil with phase-lead.
I will order new IGBTs tommorow, so i have 1 day to decide which IGBTs to use.
Simply thanks to the fact that their datasheet says "20-50kHz switching" i assume that 170-200kHz softswitching should not be a problem.
But please correct me if i'm wrong

Nice printed frame/enclosure!  Mine is built into milk crates:)  Your primary coil (and bottom of secondary) is much closer to you electronics than I'd be comfortable with.  It may be fine, but it can be hard to figure out where ~1kV/turn primary field may couple into control circuitry.  Hopefully other people will have opinions on whether this is an issue.  Perhaps I'm just paranoid of difficult-to-quantify possible issues.

Thank you Yes you're right, the primary is very close to the IGBTs and GDT. But otherwise the low-voltage PSU and the driver are (i think) well isolated. They are placed in this metal box, which is of course grounded to earth.


It hurts me to see my coil in this dissasembled state But at least all these failures make the finished product reliable.

[Hydron]

Sorry to say but the layout shown has a lot of inductance, especially within in the pairs of IGBTs within each half bridge.

It is best to put the High and Low side parts as close as possible, as the connection between them is critical for low spikes during switching. This connection should be as short and wide as possible, whereas on your design it goes all the way under the C1 and C2 capacitors and is a relatively skinny track.

The distance between each half bridge is less critical as long as there are good close connections to the film capacitors across the DC bus.

I would try and do a new layout with the IGBTs within each half bridge much closer, and much bigger tracks for all the power traces, using planes where possible (especially useful for the power rails - have one for +ve, one for -ve, one above the other for very low inductance).

[davekni]

Looks like a good metal enclosure for control electronics, and twisted pairs for all gate-drive signals.  That makes control disruption less likely an explanation for the failure.

The gate-drive layout is ideal Kelvin connection, to the extent ideal is possible with three-lead IGBTs.  No issue there.

Concerning inductance:  Trace thickness (added solder or copper) has almost no effect on inductance.  Focus instead on the open area between traces where magnetic fields are free to circulate around the traces.  Changing traces to overlapping planes (VBus+ on one side and VBus- on the other side) will cut inductance drastically, perhaps to 1/10th.  For power buses, think about full flood of copper on both sides with narrow gaps where needed to isolate different signals.  It's the reverse of thinking about traces in an empty field.  Where gaps in planes are needed, avoid lining up a top-side gap with a bottom-side gap, as that makes a place for magnetic field to sneak through more easily.

The majority of that ECB could easily have traces expanded into planes - through the bulk caps and lower two film caps.  The area between IGBTs is a bit more tricky.  Horizontal connections for H-Bridge output should also be wide, as their inductance is in the loop from VBus+ to VBus-.  If one H-Bridge output connection could be moved to just below the IGBTs, that would make low-inductance layout fairly easy without changing any component locations.

This ECB looks familiar.  Is it getting used widely here?  If there's enough interest, perhaps later this year I could lay out a low-inductance version to post, at least as an example of technique.

Concerning IXYB82N120C3H1, that's probably fine given that you have sufficient phase-lead.  IGBT diode speed often doesn't track IGBT speed very well.  Some designs (including DRSSTC with good phase lead) don't need fast diodes.  Hard-switching designs such as a typical H-Bridge motor controller do need fast diodes for fast switching.  Specifications aren't standardized well either.  Your existing part and new one specify diode characteristics quite differently.  The only point of direct comparison is Trr, 420ns for the new part compared to 41 and 126ns for the old part.  Test conditions are quite different too.

Even if phase lead occasionally isn't quite enough, 1200Vce should protect you from any resulting voltage spikes.  So, that IXYB82N120C3H1 part is probably a good choice, and will likely work with the existing ECB layout.

BTW, I'm in the same phase of design.  Tested my upgraded isolated-PFC at 8kW for the first time yesterday.  Worked for a while, then made smoke.  Paralleled output rectifier diodes didn't share current evenly enough.  Today I'll be ordering some SiC diodes that share better.

[AstRii]

Thank you for your advice!
So far i see i was too confident about my layout, nevertheless i want to keep the layout since i want this coil to be finally finished and my 3D printer is currently not working to print new elements for new layout and new PCB.
So the only change will be new IGBT (this time at least 900V, that should keep me safe from any overvoltage spikes even with bad inductance) and new at least 800V TVS diodes.
Now i'm thinking which IGBT should i use.
I was thinking about the already mentioned: IXYB82N120C3H1 http://www.farnell.com/datasheets/2173925.pdf?_ga=2.244592199.332396150.1598123582-1311482206.1541981099
Or: FGY75T120SQDN http://www.farnell.com/datasheets/2711364.pdf?_ga=2.73621521.228936157.1598203358-1258870554.1585998574&_gac=1.215291365.1596029163.Cj0KCQjwvIT5BRCqARIsAAwwD-QDlyQSu9_ZPGaPGtJMMpvRMbzvLRSR5Lcge-FOFuVHV6eRXD4JF8YaAkIXEALw_wcB
which is basically the same FGY75 IGBT as i had now but with slighly higher current rating and twice the voltage rating (1200V).
FGY75T120SQDN has faster internal diode and it's twice cheaper, but is has very long t_off and 3x the gate capacitance.

I'm targeting 300-350A OCD, 340V DC bus, 60-80us ontime, ~200kHz switching.
What do you guys think, which one sounds better to you?

Thank you,
- Mark

[davekni]

Either 1200V IGBT is probably fine.  Think I'd choose FGY75T120SQDN.  Biggest down-side is higher gate charge, but UD2.7 should handle that fine.  UD2.7 is often used with much larger IGBT bricks.  (Do check gate waveforms and lower gate resistance as needed to handle higher capacitance.)  I suspect that the switching speed differences are largely due to differences in test conditions rather than actual part differences.  FGY75T120SQDN speeds are listed for 10-ohm gate drive, while IXYB82N120C3H1 lists speeds with 2-ohm gate drive.  Gate drive resistance makes a huge difference.

The FGY75T120SQDN diode is slower than it's 600V counterpart, but faster than the IXYS part, and listed as soft-recovery.  No specific soft-recovery spec's (hard to quantify), but soft means that the recovery current doesn't end abruptly.  Soft recovery is a help in minimizing voltage spikes if you do get a phase-lag situation momentarily.  The FGY75T120SQDN also has lower Vce at high current, so should be a bit more efficient.  (Diode forward drop is higher, though, which is typical for faster diodes.  Diode drop matters only during ring-down after enable pulse end, so isn't as important.)

Good luck with whichever part you pick!

[AstRii]

Alright, so far i'm going for FGY75T120SQDN, if by tommorow afternoon i won't find better option, i will order them
Thank you!

[Zipdox]

Can you use a different image host? Facebook container is blocking the images from loading.

[AstRii]

It should be okay now, i edited the posts and used different image host

[Teravolt]

do you think that you may have any shoot through on your mosfets transitioning times because that can make for bad noisepspike roble ms. the failure mode of TVS's are to short. problems at lower voltages may not be as visible at running voltages. have you used a isolation transformer and scope to look at your bridge

[AstRii]

I'm not sure if there are any crossconduction problems.
I only scoped the bridge while powering the coil from 80V DC isolated power supply and i've meassured this.



Red waveform is the primary current and yellow is the output voltage of the inverter with 100V peaks.
I'm assuming the voltage spikes will get lower and lower at higher input voltages as IGBT's C-E capacitance decreases (but then nothing explains the destruction of the 440V TVS)

I have recently bought an handheld battery powered 2 channel oscilloscope, i will do way more measurements this time

[davekni]

Was that scope shot from before increasing Vge and squaring-up the Vge waveform, or after?  I'm trying to speculate about what causes noise in the waveforms shortly before the output step.  Perhaps that's initial IGBT turn-off followed by base storage time before switching.

What value film capacitors are you using?  The ~1MHz ring looks mostly like film-to-bulk cap wiring inductance, but is higher frequency that I would have expected.  If it is as I'm guessing, that ring amplitude will roughly track current, with bus voltage having little effect.  If this is +-140A, that ring will be less than double at +-250A OCD.  (That is double in volts, so won't look too bad on top of 340V normal square-wave drive waveform.)  The shorter switching spikes do depend more on IGBT capacitance, which changes with voltage as you know.

[AstRii]

I'm not sure what do you mean by increasing Vge and squaring-up the Vge waveform.
Last time i scoped Vge it looked like this



Red waveform being one of the low side IGBT G-E and yellow being the inverter output.
Though scoped at only 20 or 30V DC input voltage.

I'm using 4x 82n 1kV in parallel for total of 328n snubber capacitance made of Tesla TC343 impulse capacitors.
I don't know anymore how many Amps/div are there on the scope, but i think for 80V DC input, the current was about ~100Apk.

[davekni]

Sorry, I'm mixing up threads.  Please ignore my question.

[AstRii]

Hello
I've upgraded the bridge, new IGBTs: FGY75T120SQDN and instead of one 440V TVS diode i use two in series for 880V.
As i promised, this time i will scope everything i can imagine, so here are 13 screenshots of different waveforms
(for now tested at only low voltage, 20V input from lab bench power supply)



























I know the low voltage testing is not saying much as the C-E capacitance is very high, but at least something
I hate how noisy some of those waveforms are, but that's probably just by my poor scoping.



The test are done with secondary in place (i made it too permanent )

I would be grateful for any opinions, i'm pretty much satisfied with it, only thing i hate about it is the noise on the primary current, which i hope is just the switching noise, but i'm scared that the noise is there because of some current shoot-through.
Also the gate voltage could be about 24V and not almost 30V, but the IGBTs should handle that fine according to the datasheet

[davekni]

May be partially an artifact of running at low voltage/power, but the scope traces show running with net phase-lag (insufficient phase-lead).  This is easiest to see in your fourth image of Vce.  Vce goes negative for ~300ns before switching.  Negative Vce is diode conduction, caused by current reversing before gate switching.

Are the zoomed-in images taken near the end of the enable burst where primary current is highest?  If you can supply a bit more bus voltage, say 40V instead of 20V, waveforms may be more representative of normal use.

Gate drive generally looks good, with about 150ns of dead-time, so no shoot-through.  However, I don't have a good explanation for the ripple on top.  Must be some interaction of capacitive elements in the driver and GDT inductance.  Do you have any close-up images of your GDT and cabling to and from it?  Did you make any part substitutions for the gate-drivers and R/C parts in the UD2.7 output section?  What cores are you using for GDT and CTs?

One reason for possible concern about GDT leakage-inductance and UD2.7 output section is the gate waveform at the end of the enable pulse.  The IGBTs that are off just as enable ends turn on very briefly with a ~8V gate overshoot, making a glitch in Vce followed by a spike.  I don't know how common this is with UD2.7 or other GDT designs or just how problematic it is.  I'd suggest scope shots zoomed in to that final IGBT gate transition.  Hopefully lower GDT leakage inductance can limit that final gate spike below Vge threshold voltage.

The current sensing noise bursts aren't real current noise, but rather voltage noise on the H-Bridge output capacitively-coupling through the current-sense transformer to the scope.  The noise is still useful as an indication of high-frequency transients.  The first noise burst appears to start when the conducting IGBTs change to diode-conduction (reverse Vce).  That's odd, since the current waveform appears to not have changed polarity quite yet.  There's some issue with your CT and burden resistor for scoping that's showing a phase-lagged version of current.  The IGBT diode signal is an accurate indication of when current changes polarity.  More details about your CT construction and cores and what is shared among scope/feedback/OCD would be helpful.  The second noise burst is the IGBT switching, with some diode recovery spike since it is switching after diodes conduct.

BTW, one of my two previous questions is still of relevant:  What value are your VBus film capacitors?  That combined with VBus ripple frequency can help estimate VBus inductance and related voltage spikes.

[AstRii]

The zoomed-in images are not near the ends, i didn't think of that when i was scoping the waveforms. They are in about half the enable pulse interval.

The GDT can be best seen on these photos:






I did not made any substitutions to the output section of the UD2.7

I do not know which material is the GDT made of as i've pretty much took the first ferrite CT i've seen on my desk (maybe not so clever decision).

The CTs i used to scope primary current, feedback and OCD are all made of iron powder core. Works well (guessing from the fact the coil was previously working well with 300V DC bus and 200A OCD). but whenever i try to scope those signals i get this noise.
I use 1ohm burden metalic (low inductance) resistor with a ~50turns iron powder CT.

I use 4x Tesla TC343 180nF 1000V capacitors as snubber caps for total of 720nF 1000V rated 750V/us.

The CT for scoping current can be seen here:



Soon i will try higher voltage tests (i think i can go up to 80V DC with my power supplies i have here).

[AstRii]

So i've managed to scope some more waveforms this time at 50V DC bus


Yellow - Low side Collector-Emitter voltage 20V/div
Red - Primary current 25A/div
5us/div
2nd half of the enable pulse


Yellow - Low side Collector-Emitter voltage 20V/div
Red - Low side Collector-Emitter voltage 20V/div
10us/div
full enable pulse


Yellow - Low side Collector-Emitter voltage 20V/div
Red - Low side Collector-Emitter voltage 20V/div
1us/div
zoomed in into the end of enable pulse


Yellow - Low side Collector-Emitter voltage 20V/div
Red - Low side Collector-Emitter voltage 20V/div
50ns/div
close up zoom to the "normal" operation during enable pulse

[Teravolt]

I think you have transitioning noise with out using any dead time and the spikes betwean the turn on and off of each half bridge transition is cross conduction for 200-500ns. I would make dead time qbout 1us
your drive looks good but is the tesla in tune beween the primary and secondary. the secondary resonance should be slightly higher so when your arc loads it they will resonate together