GDT keeps on killing IGBTs

[ritaismyconscience]

I tried it by tying a wire around one of my primary wires. Here's some waveforms with 250V on bus:

(5us/div)




(2us/div)

Edit:
Think I found the problem:

I threaded a wire through the feedback transformer that goes to the controller, and I got this triangular waveform:





Looks like the green toroids are causing the problem.

[Hydron]

Sorry to be negative, but I think you're going to continue to struggle until:
- you get or build a driver that actually has real, open information about it's operation that you understand, and which also allows for a phase lead to be set. A random aliexpress driver might be OK if you can follow the design discussion etc on chinese forums, but if not then stick to something that's open and documented in an accessible way.
- you get some proper probes for looking at the bridge output (even a cheap ebay 100x passive probe will work if you reference it to mains earth)
- make sure that best practices are followed regarding the critical parts of the circuit. These are mainly the loop area of the IGBT connections to eachother, the output and their decoupling capacitors, GDT drive stray inductance and the IGBT gate drive network.

[ritaismyconscience]

Ok, I'll buy a cheap probe online.

In the meantime, I'm going to make the following changes:
1. Reduce GDT from 16 turns to 12 turns to reduce leakage inductance
2. Rewind feedback transformer (think the core is saturating)

My guess is that the feedback transformer is doing some phase shifting on the feedback signal. What type of core should I use for the feedback?'

Also, it seems like I just need an inductor and a resistor to adjust the phase lead.



Edit:
Someone's already made a version with phase lead

[johnf]

Leakage inductance has very little to do with turns
It is a measure of winding coupling (k)
This has more to do with core shape, winding window utilisation, and winding order.

In fact reducung turns tends to raise leakage inductance percentage wise
One thing you can do is reduce your in and out wire length of the transformer as this directly adds to leakage inductance, you need to aim for lead lengths of a FEW cm's max

[davekni]

A real scope probe would be helpful, although your H-Bridge output waveform looks fine with the simple wrapped-wire probe.  (Fine except for the 60 degree phase lag instead of ~10 degree lead that would be ideal.)

Scoping the actual feedback voltage at the driver input would be more helpful than a loop through the transformer.  It should be fine to ground the driver board to your scope.  I think the spec you found said mounting holes are driver ground.  I'm guessing that the waveform you are seeing now isn't saturation, but rather due to the +-3V clamp load that the driver board applies to the current-sense signal.

[Deleted incorrect statement about 90 degree phase shift.]  Scoping the feedback input directly will help determine what's actually going on.

Current-sense issues could be due to low inductance rather than saturation.  There's lots of info on this forum about core requirements and options.  (If your scoping current transformer is low-inductance, then your phase lag issue may not be as bad as it looks.  A way to double-check your current waveform is to scope the high-voltage side of the MMC, where it wires to the primary coil.  Use the same technique of small capacitor coupling to your scope, but the capacitor needs to handle10-20kV.  Insert some wire to the inside bottom of a glass bottle or jar, with the outside bottom placed on the MMC-to-primary connection should be safe enough.  Voltage there will be 90 degrees out-of-phase with current, plus small voltage steps at the H-Bridge output switching times.)

[ritaismyconscience]

I'm not seeing the 90 degrees phase shift...



I'll try using larger cores to see if they make a difference.

[davekni]

To make any useful comments, I'd need to know the inductances of each wniding and the coupling factors used in your simulation.  I suspect some value(s) don't match normal DRSSTC implementations.

[ritaismyconscience]

I changed the cores:



Here's what the feedback signal actually looks like. The clipped sine wave is the FB signal and the square wave is output.





(1st is 1us, 2nd is 2us)

[johnf]

Why guess when you can measure!!!!!!
Measure your gate drive transformer inductance preferably at a frequency near to what you are using then short the other windings and measure inductance again shorted over open circuit value x 100 = coupling factor. the shorted inductance figure is your leakage inductance. Most simple LCR meters have a series of fixed measurement freqs. This is a bit of kit you must have long before an oscilloscope. without an LCR meter it you may as well put all your components in a box and shake it for a random solution

[davekni]

I was mixed up about the phase-shift implications of the current-sense transformer feeding a diode-clamp load instead of a resistor load, so deleted that bit of my earlier message.  However, that leaves a puzzle as to how this driver board would ever include phase-lead.  Either it just doens't support phase-lead, or the FPGA code is supposed to be implementing some sort of PLL internally to adjust phase.  If the latter, it doesn't appear to be working correctly.

Concerning inductance measurement, you can get that by scoping the ring-down waveform when a charged capacitor is discharged into the coil.  With enough voltage on the capacitor, saturation current can be measured too.  That's what I do at home, as I don't have an LRC meter either.  (Have a very nice one at work, so I do measure parts there.)

[ritaismyconscience]

Looks like my driver simply doesn't have phase lead.

After upgrading the FB transformer, I get much better waveforms. It's switching the transistors 500ns too late instead of 1.2us, like last time. 500ns seems more reasonable, considering my IGBTs have switching times around 400ns. However, the driver is still switching the IGBTS too late. I'll probalby have to manually add a phase lead circuit somehow to solve this problem.

Here's some scope pictures.





1us/div on top, 2us on bottom

[davekni]

If the current-transformer used for scoping has low inductance, you may be even closer to good phasing.  Low-inductance scope feedback will make the current waveform earlier than it really is, so make the H-Bridge output look later relative to current.  That's one advantage of scoping (through a small HV capacitor as I've mentioned before) the MMC output.  That waveform will be 90-degrees from current, and include small (relatively) steps at the H-Bridge switching points.  Phasing can be measured with that single waveform.

Do you know if your cores are all ferrite and not powdered iron? Powdered iron cores have much lower inductance per turn-squared.  They are intended for inductors, not for transformers.

[ritaismyconscience]

All the cores I'm using are ferrite.

I tried your idea, and it seems like the driver is still switching too late. It's about 400ns off.




1us top, 2us bottom

I'll try using a couple resistors and a variable inductor to try to change the phase lead.

[davekni]

Thank you for the scope traces.  Looks to me that about 800ns of phase lead is needed, to get switching just slightly before the zero-current points.

[ritaismyconscience]

Did a small test using 2 100ohm resistors and 1 inductor (don't know the inductance, will have to measure)





Edit: the inductance is 0.38mH

[davekni]

Looks like you are getting reasonable phase lead with that setup.  However, I'm quite confused by the image.  It looks like the current transformer secondary is wired between the shields of two different coax cables, with the coax center wires just joined together.  I would have expected a single coax cable with the current transformer wired between the center and shield wires.

[ritaismyconscience]

That's so I could look at the FB input on my scope.

What's the ideal amount of lead?

Edit: Here's the best result I have currently:



Seems to have gotten rid of all the ringing (will have to verify once I get my 10x probe)

[davekni]

Reduced ringing is one way to tell that the phase-lead is sufficient.  When the phase is lagging, the current reverses before the voltage.  In that case, current through conducting IGBTs has transferred to their internal anti-parallel diodes.  When the opposite IGBTs turn on, that causes a huge current spike due to the reverse-recovery charge of those diodes that must turn-off rapidly.

With phase lead, the conducting IGBTs turn off while still conducting a small amount of current in their normal forward direction.  That small current causes the voltage to switch softly just before the opposite IGBTs turn on.  (The reason for diodes in the gate-drive is to make sure the turn-off happens before the opposing turn-on.)

[ritaismyconscience]

Here's what my coil looks like so far:

(I didn't have polyurethane varnish so I used acrylic paint instead)



Here's my topload, which I made from the limited resources I have.

The resonant frequency is around 195kHz (with a 3ft wire attached). Currently, I have the OCD set to 500A. I just need to 3D print some parts (to connect the secondary to the topload), then I can test.

Primary is about 4 and a third turns of litz wire.

[davekni]

Are those steel screws holding down the yellow secondary holder?  Steel parts in or near the primary coil will be induction-heated, perhaps enough to become a problem (melt or burn parts).

Is the bottom of the secondary winding roughly at the top of the yellow holder?  If so, having your secondary bottom above the primary like that will reduce coupling factor, which may reduce performance.  It's safer, however, in avoiding arcs between secondary and primary, which would damage the secondary.  So, perhaps it's a good way to start.  (BTW, I started with my secondary bottom 50mm below my primary.  That resulted in arcs to my grounded guard ring, requiring patching holes in secondary coating.  Raised the secondary to 25mm below primary and added many more layers of polyurethane to the secondary to fix things.)

Was 195kHz with the top-load and 3' wire, or just the wire?

Good luck!