Odin The All-Fragger: a large DRSSTC / Projects / Forums

Forums

4hv.org :: Forums :: Projects

« Previous topic | Next topic »

Odin The All-Fragger: a large DRSSTC

first previous 18 19 20 21 next last

Move Thread

LAN_403

Steve Conner

Wed Jun 04 2014, 01:09PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Yes, it stuck out because it was the opposite of what I expected. I will have to investigate more, don't go tearing up all the theory in the meantime.

I also plan to try a larger topload, which will definitely need more primary capacitance.

Steve Conner

Sun Jun 15 2014, 07:00PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Another update: One of my SKM300GB12T4s has blown out.

I'm going back to the lab to investigate the reasons for failure, here are the main possibilities I'm considering.

1) Hard switching- The PLL driver can't provide perfect soft switching at all times, because there is always a delay in tracking frequency changes due to the loop filter. I am pretty sure that any hard switching that might happen is well within the device's switching SOA, though.

2) Desaturation- The Semikron application notes suggest that the T4 chips start to run out of steam around 3x the nominal collector current rating. This is really not that much greater than the peak pulsed rating (Icrm) however they do specify it at 15V gate voltage, whereas I'm using 24.

3) General lack of beefiness- The T4 series are 2 generations newer than the good old Powerex H series bricks. They use about half the die area and run at twice the current density. Maybe there is just less headroom for abusing the maximum ratings.

4) Too fast switching. The T4 series are also about 3x faster than the H series. That means bigger voltage spikes from stray inductance and freewheel diode recovery. However I think this is low down the list as Semikron presumably redesigned the rest of the module to cope with the extra speed and tested that it could hard switch currents up to Icrm. It would take a huge phase error to push the hard switching current over Icrm and this would have been obvious on the scope during the earlier tests. Also my DC bus voltage is right at the low end of what 1200V devices are designed to handle, leaving plenty of headroom for spikes. And, the spikes would have appeared at the brick's AC output terminal, which I was monitoring with the scope.

Ash Small

Sun Jun 15 2014, 11:23PM

Registered Member #3414 Joined: Sun Nov 14 2010, 05:05PM
Location: UK
Posts: 4245

Not sure if this is of any relevance, Steve. This is from an ON semi app. note:

"The transconductance also varies from device-to-device.
This parameter is defined as the change in collector current
for the change in gate voltage. It is by no means a constant
and a typical curve is normally shown in the data sheet."

Steve Conner

Tue Jun 17 2014, 11:14AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Ash: Yes, it is sort of relevant.

Update: A box of surplus CM300DY-24H and SKM400GB123D just arrived in the mail. The sender also included a Steve Ward driver board, I wonder if this is some sort of hint?

Avalanche

Tue Jun 17 2014, 11:58AM

Registered Member #103 Joined: Thu Feb 09 2006, 08:16PM
Location: Derby, UK
Posts: 845

I've sometimes wondered if it would be necessary to view the PLL phase error, and use a more complex loop filter built around an opamp integrator, with derivative added in the form of series capacitance (to speed up the response to sudden frequency jumps).

Characterisation of switching SOAs (for a given layout) is usually done with the 'double pulse test' from what I've observed, so that switching speed can be optimised (against everything else). This would allow you to know the maximum currrent that can be turned off with a given gate resistor value, or translate that to maximum phase error somehow (and shut down before the error gets that high). I suppose it could easily get overly complicated, you might have to implement some kind of soft turn-off when shutting down the IGBTs when there is a phase error and they would have to hard-switch

Just thinking out loud... probably mostly impractical in reality. Maybe bigger IGBTs are the way to go

Steve Conner

Tue Jun 17 2014, 01:16PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

No matter how complicated a loop filter you dream up, you can dream up some scenario that will defeat it.

The primary current of the Tesla coil can even contain more than one frequency, and no PLL can handle this as the VCO can only produce one frequency at a time.

Derivative gain will amplify ripple and lead to the duty cycle of the VCO output going uneven, which will cause hard switching.

You have to set off the disadvantage of the PLL (always some hard switching, but can quite probably be kept within the switching SOA) against the advantages (operation on upper pole frequency with primary tuned below secondary, very easy to implement pulse-skipping current limit, keeps running even for low values of loaded Q that might stop a feedback driver)

Steve Conner

Wed Jul 09 2014, 10:10AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Update July 2014: (Edit: removed comment about failed test run as I already covered it in a previous post)

I set the system back up in the lab, as a half bridge with the remaining good IGBT. I broke the top off it so I could probe it at the die level, and I also added a small Rogowski coil around the emitter terminal. The following scope shots show the collector voltage (probed as close to the die as possible) and the Rogowski coil signal showing the emitter current of the same device. The PLL driver settings were as close as possible to the last fatal run, I just adjusted the frequency to account for the fact that it was running with no secondary coil. The DC bus voltage was reduced as I didn't really want a face full of IGBT goop if anything went wrong. :)

These results suggest that the PLL driver was set up with too much phase lead. The funny thing is that even with this large hard switched current, and turning off the IGBT as fast as possible with no gate resistors, and probing as close to the die as possible, the voltage waveform is very clean. I can't see any spikes that would endanger the breakdown voltage of the device. If I tune the phase lead for perfect ZCS according to the Rogowski coil signal, the spikes get somewhat worse, and if I tune it for leading current, the spikes get terrible.

My next move will be to set it up with SKM400GB123Ds, gate resistors, current limit increased to 1kA, and the PLL retuned for less hard switching. I will try to log the Rogowski coil waveforms while making sparks, and that should settle once and for all whether the PLL driver is good enough for the job.

1404900606 30 FT4896 10496059 10152608278251584 2505759111616666150 O

1404900606 30 FT4896 10454371 10152608278231584 3861835156770858505 O

Hydron

Wed Jul 09 2014, 12:46PM

Registered Member #30656 Joined: Tue Jul 30 2013, 02:40AM
Location: UK
Posts: 208

I have also noticed that the spikes seem to be a little smaller when phase lead is set a bit too high. I didn't test with a rogowski coil so I'm at the mercy of any small phase error of my CT in judging the exact zero crossing, but I could see the diode start conducting in the voltage waveform when I'd adjusted for minimum voltage spikes. Can't say much more right now, in Asia on holiday so this is all from memory.

Goodchild

Wed Jul 09 2014, 02:27PM

Registered Member #2292 Joined: Fri Aug 14 2009, 05:33PM
Location: The Wild West AKA Arizona
Posts: 795

Steve, this is just my opinion. I have seen a lot of other systems run with the same IGBTs, and similar secondary and primary capacitor setups with little issue up at 1.2kA with the SKM200s. I personally have done this, without a single failure (as of yet). I donâ€™t think your trouble has been solely caused by the wrong secondary/primary configuration.

I think that you really need to take a hard look at your controller and gate drive setup. I strongly suspect that the reason you are detonating IGBTs is the fact that somehow the switches are hard switching or possibly partly operating in the linear area and not in the SOA.

If you have time, it may be worth your while to try a direct feedback driver with no PLL and/or direct GDT operation. I can send you one of my driver boards if you like, purely to see the different in operation. I know you swear by your PLL driver; however in the spirit of experimentation it may be worth it to weed it out as a potential problem causer. If it turns out operation is just as pore then you have simply eliminated it as the problem, no harm done.

The larger topload also sounds like a good addition.

Thatâ€™s my two bits at least.

Steve Conner

Wed Jul 09 2014, 03:05PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

I asked about this in a recent thread. The only person who posted any relevant results for the exact devices I destroyed was Kizmo who blew up a 128 series at about 1500A, a similar current to what caused my 128 series to fail. Most of the Semikron experience seems to be with the 123 and 124 series. The SKM300GB12T4 has no track record whatsoever, and only about 1/2 the die area of the devices normally used in Tesla coiling.

Also, from the previous post you can see that I am taking a hard look at my controller and gate drive setup, and it showed that there probably was excessive hard switching going on at the time of the failure.

This next run with the PLL driver will be its last chance. If it still causes excessive hard switching under real streamer load conditions, this will be obvious from the Rogowski coil signal. In this case I will replace the PLL driver with a feedback driver. An anonymous donor already sent me some of Steve Ward's phase lead driver boards as some kind of hint.

I would be interested to see the Rogowski coil signal from a big DRSSTC that is known to work reliably.

first previous 18 19 20 21 next last

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob