MMC capacitor crash

[Mads Barnkob]

SDS00004.png shows a normal classic voltage transient inverter situation where current ramps too high and the DC bus is pulled down, which then limits current and DC bus crawls back up again as the MMC from from empty (a short circuit to the inverter) to less load and current falls. As Dave says, this is a sign of no energy transfer.

random example from the Internet(tm)


However, the latest SDS00155.jpg shows that you have a ringup of about 200 uS, 200 us additional where it falls abit and goes into steady state for 600+ us. The steady state shows a continues energy transfer to the secondary circuit, energy fed into the inverter goes all the way to the spark. It does not help the spark growth, it just keeps it on for a longer time. Your spark growth is all in the ringup.

[davekni]

here generator input feedback f = 153khz

As Mads has pointed out in other posts, running open-loop can lead to IGBT failures due to excess hard-switching (IGBTs switching at high current parts of the waveform).  Certainly keep Vbus low as you have been for such testing.

I'm guessing that SDS00153.jpg is slightly below primary resonant frequency.  Running open-loop (from signal generator) below resonance causes phase lag.  That is the most risky (for IGBT failure) condition where turn-on is after current reverses.  Resulting switching spikes are clipped in scope capture, presumably because your differential probe is still set for 50x rather than 500x.

why this change in length ? instability ? vbus too low,

All three traces show operation longer than 200us.  If your interrupter is set for 200us, either your interrupter or driver board is not functioning correctly.  I suggest finding and fixing that issue first.  If pulses are lasting much longer than interrupter is requesting, that could have been the initial cause of MMC crash, much higher duty cycles than you intended.

SDS00004.png shows a normal classic voltage transient inverter situation where current ramps too high and the DC bus is pulled down, which then limits current and DC bus crawls back up again as the MMC from from empty (a short circuit to the inverter) to less load and current falls.

With respect for Mads' experience, I think this isn't the case here.  Inductance in the differential probe connection to H-bridge output confuses the image.  The individual H-bridge output steps can still be seen, and decrease in amplitude only slightly across SDS00004.png capture.  Also, primary current rise rate decreases only slightly, indicating Vbus is still close to its initial value.  Your scope captures will be easier to understand if differential probe connection can be improved.  Can you post a picture of probe connection including probe leads?

[Mads Barnkob]

SDS00004.png shows a normal classic voltage transient inverter situation where current ramps too high and the DC bus is pulled down, which then limits current and DC bus crawls back up again as the MMC from from empty (a short circuit to the inverter) to less load and current falls.

With respect for Mads' experience, I think this isn't the case here.  Inductance in the differential probe connection to H-bridge output confuses the image.  The individual H-bridge output steps can still be seen, and decrease in amplitude only slightly across SDS00004.png capture.  Also, primary current rise rate decreases only slightly, indicating Vbus is still close to its initial value.  Your scope captures will be easier to understand if differential probe connection can be improved.  Can you post a picture of probe connection including probe leads?

It was not as based on my own experience, as it was more based on similar waveforms in inverters.

I will stand corrected if that is the case I have never seen so bad measurements with a differential probe, but then again, I assumed the measurement here is not to blame. But definately worth checking up upon with some pictures.

[JCF]

Merci pour l'aide !
here are 2 photos that show the box with primary and details of HV probes connected to the bridge output.


with toroid on feedback I see f=  151.6khz is it  necessary to adjust the frequency of the primary to 153khz. 153khz is maximum spark lenght ? ?
the instability came from the contact on the aluminum crown : it is ok now.
sense for current is 1:300 on 100 ohms.
input fiber optic signal 200µS and 10hz sqare

it seems to be working fine ?

[davekni]

here are 2 photos that show the box with primary and details of HV probes connected to the bridge output.

The unusual H-bridge output scope traces are due to inductance of wires from IGBTs to front edge of the black box.  If it isn't too hard, run a second pair of wires (may be smaller wire) from IGBTs (H-bridge output) to the back side of the black box.  Connect scope probe to those new wires instead of to wires connected to MMC and primary coil.  That will avoid most wiring inductance that is common to both primary coil path and scope path.  Scope traces will look more like square waves.  Any subtle issues will be easier to see and correctly interpret that way.

with toroid on feedback I see f=  151.6khz is it  necessary to adjust the frequency of the primary to 153khz. 153khz is maximum spark lenght ? ?

By "toroid", I'm presuming you refer to the feedback current transformer, not the top-load aluminum toroid.  If UD2.7 phase-lead is adjusted optimally, I'd expect it to oscillate at 153kHz based on previous open-loop (signal generator driven) scope traces.  However, that 151.6kHz to 153kHz is a tiny difference, not that significant.  For maximum spark length, primary frequency needs to be a little below secondary frequency.  Enough below so that it matches secondary frequency as secondary frequency drops due to spark loading (spark capacitance).  JaveTC is a good way to model primary and secondary frequencies (and coupling factor etc.).  Include wire length in JavaTC input since your primary wire length is significant.  (In general, I'd recommend pairing primary leads closer together to reduce wire inductance.  That will increase primary frequency slightly, so might require adjusting primary coil to compensate.)

it seems to be working fine ?

Looks fine in that pulses last 200us as intended.  Still don't see as much evidence of power transfer to secondary as shows up in the one old scope capture SDS000004.  That makes me suspect secondary frequency isn't close enough to 153kHz in this test.

Merci pour l'aide !

You are certainly welcome!  (Google translate worked well on this short phrase.)

Good luck!

[JCF]

Hello, thank you for help 
I'm back with the last oscillograms after adjusting the primary at 153kHz: primary winding raised by 1 cm towards the secondary in the lower part, and removed 1/2 turn from the primary
here is the 3d view of the coil and the calculation with javatc3d1.8

SPEC Coil 18 70 2900 V2.txt
The primary and secondary frequencies are very close: about 2/3Khz. is it too close, what are the risks see SDS00003

from David :  Looks fine in that pulses last 200us as intended.  Still don't see as much evidence of power transfer to secondary as shows up in the one old scope capture SDS000004.  That makes me suspect secondary frequency isn't close enough to 153kHz in this test.

I don't understand this sentence, how see evidence of power transfer to secondary ?

on the trace of the signal SDS00001, why are the square signals at the beginning transformed and become species of sawtooth the closer we go to the end of the signal (around 200µS) ?
 the sinusoidal signal measured via the current probe (1:300) has a voltage of 74V DC on 100 Ohms, this gives 220A C/C? I find it huge for about 100v C/C on the primary, what do you think about?

should any other changes be made?

[davekni]

The primary and secondary frequencies are very close: about 2/3Khz. is it too close, what are the risks see SDS00003

Your JavaTC text lists primary resonant frequency at 154kHz and secondary at 134kHz.  Primary frequency should be a little below secondary frequency, not 13% above.  That may explain the relatively-little energy transfer from primary to secondary.  Coupling factor of 0.161 should be fine.

I don't understand this sentence, how see the secondary evidence of power transfer to secondary ?

Power transfer shows up as amplitude modulation in primary current waveform and/or a plateau in primary current if secondary arcs lower Q sufficiently.  Latest graph shows a hint of these effects.

on the trace of the signal SDS00001, why are the square signals at the beginning transformed and become species of sawtooth the closer we go to the end of the signal (around 200µS) ?

The sawtooth part is due to remaining common inductance of H-bridge between scope wires and output wires (wires to MMC and primary coil).  Sawtooth amplitude is about 1/3rd of what it had been previously.  Nice improvement.  Low enough sawtooth amplitude to allow looking past that to see square-wave part.  (Mentally subtract out that sawtooth part of H-bridge output waveforms.)

the sinusoidal signal measured via the current probe (1:300) has a voltage of 74V DC on 100 Ohms, this gives 220A C/C? I find it huge for about 100v C/C on the primary, what do you think about?

Once primary frequency is reduced to slightly below secondary frequency, you will get more power transfer to secondary, reducing primary current.

should any other changes be made?

Just fix (reduce) primary frequency.  Increase primary turns and/or increase MMC capacitance.  If you have a signal generator, you can measure secondary frequency to verify JavaTC's 134kHz estimate.

[JCF]

well, now primary with 15 spires
 
with generator on feedback137KHz see picture SDS0007


with feedback, what is the trouble ? see picture SDS0006


javaTC Javatc3d.txt SPEC

[Mads Barnkob]

Do you have a 10k power resistor that you could try to slap on the inverter output? There can be some really weird behavior from the driver if your MMC has residual charge or a DC offset from the last pulse.

The 10k power resistor just needs a power rating according to your DC bus voltage and the 10k resistance.

[davekni]

with feedback, what is the trouble ? see picture SDS0006

Current waveform is exactly what you should get.  Primary and secondary frequencies are the same.  Energy transfers from primary to secondary and then back to primary.  This is good, not a problem.  You will want to tune primary a little bit lower eventually to match the lower secondary frequency caused by arc capacitance.  During initial testing, I'd suggest leaving primary frequency as you have it now, exactly matching secondary frequency.

with generator on feedback137KHz see picture SDS0007

Looks like 137kHz isn't exactly at resonance, perhaps a bit higher than resonant frequency.  Since closed loop (with feedback) is working correctly, no need to continue testing open-loop (signal generator on feedback).

There is very large coupling of extraneous signal into H-bridge output waveform.  Appears to be more from secondary current and/or voltage than from primary current as previously.  (Possible that previous coupling was also from secondary, and I was guessing its source incorrectly as being primary current.)  Perhaps the differential probe electronics are not shielded well (especially if a cheap probe version) or there is a bad ground connection between probe and scope or   Makes interpreting H-bridge output signal almost impossible.  You could try wrapping the differential probe housing in foil and grounding that foil to scope ground.

[Anders Mikkelsen]

Do you have a picture showing the construction of your H-bridge?

[JCF]

Thanks for the help.

here are 2 more correct measurements of the day, I think?

For the first I am amazed by the resonance frequency of around 120KHz and 110 KHz photo SDS0001 (measurements made yesterday 132KHz ? to be continued...)

Current feedback 222A

I will put a resistor of 100KHz at the output of the bridge
same conditions 10Hz,200µS, 60/80V DC

[davekni]

For the first I am amazed by the resonance frequency of around 120KHz and 110 KHz photo SDS0001 (measurements made yesterday 132KHz ? to be continued...)

Looks like 120kHz is more accurate for both H-bridge output and current.  Not sure why scope is displaying 110kHz.  BTW, SDS00004 looks to be about 136kHz.

Current feedback 222A

Is 222A the OCD setpoint?  Doesn't look like either of the scope traces show nearly that much current.  If I recall correctly, current measurement is by 300:1 CT into 100ohm burden resistor, resulting in 3A/V to scope.

I will put a resistor of 100KHz at the output of the bridge

Good idea, as Mads said.  I have not seen any evidence yet of issues caused by a lack of a resistor in your scope traces.  Good to make sure with a resistor just in case.

Looks good at this point.

[JCF]

Hello friend specialists ! 
I made a calculation with the MMC calculator, looked for the specifications of the Cornell Dubilier 940 3000V 47nF
https://kaizerpowerelectronics.dk/calculators/mmc-calculator/
the 2 IGBTs in full bridge have a maximum current of 200A
By looking for frequency pairs 100bps and on time 350*S = that's not much ? d'augmenter le on time jusqu a 1ms. peak MMC current 200A MMC rating 204 A
is it possible to to increase the on time up to 1ms ?
when I change the on time to 1000µS, nothing changes in the current peak maxi area I had trouble finding which areas were calculated, I may have overwritten a formula ?

MMC calculator 0.047µF.pdf
MMC calculator 0.047µF.pdf

[davekni]

By looking for frequency pairs 100bps and on time 350*S = that's not much ? d'augmenter le on time jusqu a 1ms. peak MMC current 200A MMC rating 204 A
is it possible to to increase the on time up to 1ms ?
when I change the on time to 1000µS, nothing changes in the current peak maxi area I had trouble finding which areas were calculated, I may have overwritten a formula ?

MMC specification of most concern for your coil is RMS current, which depends on duty cycle (on time * BPS) and RMS current during bursts.  Peak does not change with on-time.  You've listed peak as being 200A.

The unknown parameter now is RMS current during burst.  If peak current is close to 200A for most of the burst, then burst RMS current is 200A/sqrt(2) = 141A.  If current only occasionally hits 200A and is lower for most of each burst, then burst RMS current will be lower too.  I'll use 141A RMS during bursts for now, as a worst-case.

Your MMC is rated for 17.1A (5.7A per capacitor) RMS.  This can probably be exceeded a bit, and a bit more if you add a fan to cool MMC.
MMC RMS current is burst RMS current * sqrt(duty cycle).  For 100Hz 350us, 141A * sqrt(0.035) = 26.4A.  Probably OK with fan cooling for MMC.  If actual burst RMS current is lower, then might be OK without cooling.

For 1ms bursts at 100Hz (10% duty cycle), MMC RMS = 141A * sqrt(0.1) = 44.7A.  That would likely cause MMC failure (if burst current is really 141A RMS).