QCW problem with ZVS and phase lead (solved)

[V-Troxi]

Hi everyone ,

This is my first post here (I've been mostly active in a German forum in the past) and I'm currently working on a buck-powered QCW Tesla coil.
I've already built and tested the bridge, the buck converter, the control circuitry, the ramp generator/interrupter and the primary and secondary. Or in other words: the project is almost done.

A quick test run of the DRSSTC part with no buck converter connected and around 60 Vbus resulted in 15cm sparks and verified that the coil was working as intended.
So I moved on to adjusting the phase lead... And that's where the problem lies.
I've used a 7m3-123 inductor with 9-15uH but I quickly realized, that it had not enough inductance, because switching occurred after the zero crossing of the primary current and was quite noisy too like in these pictures (please excuse the bad image quality :/):


(blue: current, yellow: bridge output)

(blue: Vce, yellow Vge)

Because of stupidity, I've just bought one inductor, but luckily I found an inductor with a plastic housing and an adjustable ferrite core from a broken CB radio:


Then I began to wind and rewind this inductor but I could not archive ZVS (at some point I came close to ZCS but I prefer to have ZVS because I'm building a QCW with long on times).

No matter what inductance I tried (from 5 up to around 50uH) the signals looked just weird...
I tried to use bigger inductances to shift the switching slightly before zero current (hoping that Vce would shift too and fall to 0V before the gate voltage is too high for clean ZVS), but doing this resulted in these ugly signals of Vce and the inverter output and I don't really get why:


(blue: current, yellow: bridge output)

(blue: Vce, yellow Vge)

In the last evenings, I've tried countless different phase-lead inductances but I have not archived ZVS once and anything apart from the ZCS resulted in voltage spikes and weird switching like in the previous pictures.
I am starting to doubt that the inductor is the problem.
Maybe I have overlooked something crucial during these tests or there is something severely wrong with my design...

At some point I've also burnt R2 on the UD2.7 because the inductor connection was loose and resulted in overvoltage - after replacing it, the driver worked again and I don't think the comparator got damaged.

Here are the specifications of my system:

Driver: UD2.7C
Cbus = 24000uF @ 500V (the tests were made at 120-150V with currents of 50 to 80 Amps and an empty food can filled with a bit of water as a static load)
Bridge: IXXN110N65B4H1 full bridge with two 4.7uF snubber caps, 12 Ohms of gate resistance with antiparallel STPS 2H100 schottky diodes.
GDT: TX42/26/13-3E27 with 7 turns of a Cat5e UTP 24AWG ethernet cable
Feedback/OCD: 13x13 turns on Epcos/TDK N30 cores.
Primary frequency (during testing): ~407kHz

-> Additionally I could measure no phase shift between the feedback CT and my measuring CT (100 turns on a TX42 core with a 10 Ohm shunt)

The power electronics:




The coils with test sparks at 60V:


I hope that I am not annoying you with my question(s), but this is my first time tuning for ZVS and not ZCS...

EDIT: This is the closest I could get to ZVS using the Coilcraft inductor at a lower frequency:

I've also sampled a selection of other inductors from Coilcraft (which are in stock but expected to be shipping in December!? ), because maybe my own inductor is just behaving weird due to its core, lack of shielding or whatever...


Maybe somebody's got an idea and can help me

Best regards,

Vtroxi

[davekni]

Thank you for a great set of images and scope plots.  Saves time asking basic design questions.

12 ohms gate resistance doesn't seem too high for IXXN110N65B4H1, but it appears that dead-time between turn-off and opposite IGBT turn-on is too large for 400kHz operation.  Those triple-transitions of Vce occur when there is enough phase-lead for turn-off, but not enough for the opposite IGBT to turn-on before current changes polarity.  Looks like about 300ns excess dead-time.

One issue may be excess GDT lead inductance.  Ideally each of the four GDT primary winding leads are twisted as a pair all the way back to UD2.7 output.  Short leads help too.

Have you scoped UD2.7 outputs at the UD2.7 ECB?  Should have fairly rapid clean transitions there.

May need to reduce gate resistance if everything else is good.

Any chance the IXXN110N65B4H1 parts are counterfeit, with higher gate charge?  IXXN110N65B4H1 gate charge is lower than other IGBTs of similar current capability.

[V-Troxi]

Thanks for your kind and helpful answer!

Now that I think about it, the issue with dead time seems logical.
I am going to reduce the length of the GDT leads today and measure the UD2.7 output.

I've also not thought that 12 ohms of gate resistance would be too much, but the issue could indeed be counterfeit IGBTs with higher gate charge because I bought these for a quarter of their (Mouser) price on eBay...
I still have four of them laying around since they were so cheap. Is there a relatively quick way to measure their gate charge and see if they are fake?

To be sure I could also buy 4 of them (or even their faster version ...C4H1) on Mouser but this wouldn't be very cheap - but probably the best option to be sure 
EDIT: Yes, I don't trust my IGBTs so I ordered four IXXN110N65C4H1 on Mouser and some 8 and 10 ohm 2W gate resistors in case 12 ohms is too high.

Would you say that when the dead-time issue is resolved, I would need a higher phase-lead inductance than 15uH, or is it sufficient?

Best regards,

Vtroxi

[davekni]

Do you have a meter that can measure capacitance?  If so, start by measuring Cge with 0Vce.  (Just noticed that the spec isn't quite self-consistent.  Table lists typical Cge as 3650pF at 25Vce, while the graph shows it as ~5500pF.)

What inductance were you using for those Vce tripple-transition scope plots?  That is probably about right.

UD2.7 phase lead can never quite reach 90 degrees, which would be 614ns at 407kHz.  60 degrees is a more practical upper limit, preferably 45 degrees or less (307ns at 407kHz).  However, that leaves only 307ns for all delays (comparitor + 74HC08 + UCC27423 + gate_drive_FETs + dead-time_delay + IGBT turn-on_delay).  For high-frequency coils such as QCW, all those delays should be minimized.  The last two IGBT delays are usually the longest.  However, some are easy to improve, such as 74AC08 or ... instead of HC08.  Lower GDT leakage inductance helps too.

45 degrees is when coil reactance equals CT burden resistance (51 ohms + coil resistance).  At 407kHz, 20uH has 51 ohms reactance.  It would be great if 15uH is enough, but you may need a bit more.  (Faster circuitry needing less phase-lead compensation makes timing more stable as frequency drops due to arc loading.)

Hope your new parts work well.  Experiment with your old ones too.  Better to take risks with cheap parts.

[V-Troxi]

Hi,

Here's an update on the progress.
After receiving the new IGBTs and changing the gate resistance to 8.2 ohms I quickly achieved great looking ZVS:
These plots show the signals at 120 Vbus and 378kHz:




Vge vs Vce looks very nice and I'm also quite happy with Vout vs Iprim. Although on one hand, I am wondering If the phase shift between Vout and Iprim is a bit too much there but on the other hand, decreasing it resulted in bad ZVS and switching transients like before so it's probably good.
What do you think?

After these promising results, I connected the buck converter and slowly ramped up the voltage.
The output was stunning and even too much for operation in my basement:


(~165cm)

This was achieved at around 250V on the bus and a 20ms ramp - I can even go up to 320V but I must be careful to not go over the 200A current limit of my transistors.

In my euphoria, I got a little careless and twisted the wrong knob on my ramp generator... BANG
One transistor of the bridge exploded and even the buck switch died...

The problem: I set the quadratic compensation for bus voltage falloff to an inverse value so the voltage ramp was not rising linearly but more like an x^0.3 curve.
The resulting current increased just too quickly. In addition to that, I set my OCD to 180 amps assuming a CT ratio of 13x13 instead of the real value of 14x14. Well until the OCD reacted the current was well above 200 amps...
One transistor shorted across the bus voltage which also killed my SKM400 brick.
Just one weird thing: I'd expect a shoot-through to kill at least two transistors but it seems like only one blew up. But I could also be mistaken because I've not disassembled the bridge and checked each IGBT on its own...

I've already received the new components and I'm also equipping my buck driver with an OCD circuit similar to that of the UD2.7 before any other testing.

Here's the plan:

An LM311 compares the output current via a CT and a shunt. If it's higher than a set threshold voltage it outputs HIGH and resets a CD4043 flip-flop which in turn disables the enable input of my IXDD630MCI gate driver.
According to my example calculations, it would take around 146us until the current in the buck inductor reaches 400 amps. The shutdown process with all delays should be way faster (probably just less than a few us).

I'm also going to set the OCD of my UD to a lower value (of course using the right winding ratio  ).

Do you have any other suggestions?

PS: I'm also currently building a solid case for the coil, so I can test it outside

Best regards,

Vtroxi

[davekni]

Impressive results!  I've seen other posts here showing surprisingly different results outside, with more curved looping arcs.

Vge vs Vce looks very nice and I'm also quite happy with Vout vs Iprim. Although on one hand, I am wondering If the phase shift between Vout and Iprim is a bit too much there but on the other hand, decreasing it resulted in bad ZVS and switching transients like before so it's probably good.
What do you think?

I'd keep clean ZCS (ZVS for IGBT turn-on), even though IGBT switch-off losses are higher at higher current.  You can reduce turn-off current (reduce phase lead) if you first reduce dead-time.  Perhaps 2 ohms gate R or even 0 ohms.  Bipolar gate drive waveforms produce some dead-time just due to their non-zero slew rate even w/o gate resistance.  Looks like you still have more dead-time than needed.

One transistor shorted across the bus voltage which also killed my SKM400 brick.
Just one weird thing: I'd expect a shoot-through to kill at least two transistors but it seems like only one blew up. But I could also be mistaken because I've not disassembled the bridge and checked each IGBT on its own...

My guess is that two are blown.  IGBTs usually fry shorted, but sometimes the resulting fault current is so high that bond wires or die metalization burns open like a fuse.

[Didik8774]

hello friends, it's very interesting to learn, I'm also learning, oh yes there are some screenshots of you, I hope you can explain the oscilloscope probe point when you test, VGE.VCG.BRIDGE CURRENT AND OUTPUT?.thank you