Just to make sure I understand that concept correctly; although the RMS current going in the center of the primary is greater than what flows either side, the heat generated is less because the frequency in this part of the circuit is much less than what it is coming out of the primary, and so the effective section is section (driving the R, and heat up)?

It completely makes sense, the fact the gate resistor was a 2W actually made me wonder, particularly when I realized I didn't have one handy.
Good thing we talk about this because it reminds me of another interesting thing I picked up.

Because I didn't have a 2W resistor (but only 1/4W) I pulled the thermal camera as to not run the system too long. And well yeah these 1/4W resistors do heat up, as expected. However something else caught my attention:

The primary is wired with the positive DC coming straight into the center tap (after the inductor). As such I as expecting that wire to be hotter than the other 2 wires which should only see half the average current. However, the FLIR short revealed the exact opposite. I've attached a screenshot and you can see the 2 outer primary wires being significantly hotter than the center tap wire (it's right in between the 2 outer wires which appear yellow; note: you can see a 4th wire close to the right primary wire - that 4th wire is the secondary lead going to the ground).

That puzzled me, so I pulled the clamp meter (all the other oscilloscopes and probes were still hooked on another experiments) and found numbers that match my expectations!!! roughly 3.5A in the line going to the center tap and 1.6-17 A on both the other wires. All three are from the same spool, machine wire, 14AWG, stranded, and same length +/- 1/4".

I'm sure there's an explanation, but I didn't immediately see one. The clamp is not super accurate, I'd expect +/- 0.1A - measuring AC, which may not be appropriate, I can't tell without taking a look at the waveform first. But thermal images don't lie. These 2 outer wires were definitely hotter and I can't explain it.

On that test I was pulling around 60W, and the tank capacitor was a large 20 uF (the largest one I tried which gave similar performance to the 4.7 uF one), the shot is probably 1-2 mn after start up.

I actually started with a 3D printed housing. I made 2-3 different designs which all failed but there is quite a bit more to experiment. One of the design was pretty satisfying in regards to the cross over, basically a groove coming from the top of the previous chamber, down to the bottom of the next one. This gives the advantage of spacing out that cross over wire from other wires.

The main reason for these failures is that I wasn't using oil, the number of turns was significantly lower than what I achieved with the turned housing and thought I would get around with it. Failures were due to arc'ing to the ferrite core:
- first try was with a 2-piece housing and it went through at the silicone filled seam
- the next one was a 1-piece housing with in-fill it went through the material
- the final failed in the exact same way despite 100% infill

I am pretty sure that 3D printing is the way to go, when I get my hands on a SLA printer, I will design an open cell unit for oil filling (and with vacuum degassing), with the groove mentioned above.

Last night I designed another turned PTFE unit, much larger as I found some UY30 ferrite. It won't be as compact, but generally speaking I think there are no substitute for size when it comes to this. The bigger the transformer can be, the easier it is to prevent arc'ing.

Hi,

I am creating a new thread before splitting off too deep from the original thread (link: https://highvoltageforum.net/index.php?topic=1374.msg10493#new )

I'll re-post my last message (and David's reply) as it gave a fairly detailed description:

The transformer is fully custom, I'll post pictures later tonight. It's made of 2x "C" shape ferrite bolted together. The secondary is roughly 26mm ID x 50 mm long (that includes 15 separations that are about 2mm thick). The core has an OD of 20mm, meaning 3mm of insulating PTFE. The segmented secondary was turned on a piece of PTFE. The core is not gapped.

For the primary I have 2x 2.5 turns, which is a bit on the small side which could explain the lack of success with the ZVS driver.

After running the transformer on the SSTC driver, I decided to make my own ZVS driver. And since I ended up using roughly the same components, results were the same outside of the capacitor that was film but with high ESR, it ended up heating a lot so I replaced it with a better quality capacitor with significantly lower ESR but also slightly higher capacitance. Results were immediately better, and after playing with the few caps I had available, found that 4.7 uF was giving me the best results. Note: I did go back to the commercial ZVS and put the same 4.7uF cap across the far ends taps of the primary and results were much better, but as good as mine oddly - perhaps due to the other caps still being there. I say "caps" because that driver used 2 for some reason. The ZVS circuit I am familiar with use 1 cap across the far ends of the primary, and this one uses two. I looked at the PCB and it looks like they use 2x ~0.330F but I am not sure how it is all wired together.

I proceeded with assembly of my 200mm plasma ball, flushed it with argon at 1atm and results were great. Voltage was probably a bit high because I can't run a vacuum on that cheap plastic globe (that has a flat bottom - I tried by the way, and it's not a good idea, haha). I used a grounded striker to test contact and it would definitely shock you as opposed to normal plasma ball.

The goal is to make a bigger one (glass, round, under vacuum) so I don't think I will dial down on the voltage yet, just to make sure I have enough for the larger one down the road.


I did experience a set back with a failure of the secondary. The last chamber, last few turns (hard to tell exactly) arc'ed through the PTFE to the ferrite. Not through the thickness which was 3mm but through the last separation which was only 2mm in my design for some reason.

I've made revisions to the design and will turn it this week, increased the thicknesses separating the windings from the ferrite to a total of 4mm. I had to reduce the thickness of chamber to chamber separation to compensate, but I think this will be fine, as I don't expect any arc to get through 1mm PTFE on two adjacent chamber. If it took 16 chambers to go through 2mm or so on the HV side where the potential difference is maximum, I think I should be fine with 1mm between 2 adjacent chamber.

Another setback today after rewinding a spare V1 core. I wound this one with only 2 differences: I didn't fill the last chamber (HV side), and I tried to compensate by adding 1mm more thickness on all of the other chambers. That one failed with an arc within one chamber, along the surface of the PTFE which darkened, but no penetration. I think what happened is that my layering wasn't great on this way, I kinda rushed it last night, the added layers didn't help as it increased the voltage difference within one chamber.


With all tests I found that I was lacking a direct voltage measurement on the HV side. I actually was able to use a HV probe on my early tests (that didn't work well) and measured 8kV. After tweaking the capacitor on my own ZVS I was able to get intense plasma at ambient air and discharges over 2" long, so there is no doubt that my 10kV probe wouldn't be able to work there.

And here's David's reply:

It would be interesting to see waveforms for this - primary voltage and secondary with just antenna pickup (reasonable phase, just no amplitude calibration).  ZVS drivers usually aren't used with ungapped inductors.  Inductance is high and saturation current low.  I wonder if your system is running at a frequency determined by leakage inductance and the primary and secondary (intrawinding) capacitance.

Besides the HV end failure, the common tricky problem with segmented bobbins is the wire transition from the top of one segment to the bottom of the next.  Unless there are slots or other accommodations in the separation walls, that wire to the bottom ends up adjacent every layer (including the top) of the new segment as it is wound.  That leaves a full segment's voltage across two thicknesses of enamel insulation.

BTW, the two 0.33uF (1200Vdc, 630Vac) caps in the commercial ZVS are almost certainly in parallel.  They are standard Chinese induction cooktop capacitors, the ones I use for my DRSSTC MMC.  Some ZVS units have 6 or 8 in parallel.

And my response:

- The driver I am using is this one: https://www.amazon.com/gp/product/B07BNZ5HC1/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1

- The first HV failure was definitely a hole through the PTFE, extremely small. The second HV failure (today's failure) was exactly as you described. The wire coming from the previous chamber arc'ed with another one from the new chamber. This is a typical difficult to address mechanically. I initially tried to add a tiny hole at an angle through the separation, which was just too difficult to do without at least a 4th axis. Then I thought about just using a groove. I figured that oil fill up that groove, while this could work I think I haven't been able to machine this, it is entirely possible I just need time to actually setup in the CNC machine one of my spares, or even start with one of the damaged ones.

- I haven't hooked up the oscilloscope for relevant tracing yet, but I will do that on my next try. On one early try with the HV passive probe I measured 8kV and about 50kHz.



A few more things:

As you know me by now, I am not familiar with the technical jargon and I tend to describe things at length when in fact there is usually a simple technical term for it. I'm sorry about that; so you'll judge by my pictures but I'm pretty sure my ferrite core isn't gapped, just two "C's" forming a "O" with a centered-tap primary on one cylinder and a 3000 turns or so secondary. Secondary inductance was measured to several dozen henry, and its DC resistance to ~ 370 ohm.

I have very little footage with that first secondary design, I should have recorded more, it actually worked for a few hours before failing, it actually only failed because I hooked up a larger PSU (I was using a 60V capable PSU but limited to ~3A, so voltage input was around 13V @ 3.2A) and it failed around 17V and 10A or so. So here's the only video I have, sadly with all lights on. It doesn't show it "starting up", which at times can be tricky (I need to use a switch and perhaps use a large bulk cap). It worked pretty well overall and without the need for a sharp breakout point. Also the video was taken as I was experimenting with the Amazon ZVS with the 4.7uF which I found to be optimal with my home made ZVS. So the video shows a little bit less performance than I had with the home made ZVS.

/>
Specs:
- roughly 3000 turns of 34AWG enamel wire
- U shape cores I used: link
- a (unnecessarily) large oil container (I only had 1ft tubing and nothing to make a cut straight enough for a good seal)
- all parts for the container and globe base/globe seal were CNC'ed
- the globe itself is a plastic flat neck (not sure really the material) and sadly a flat bottom (not vacuum/pressure happy) - from amazon. Took this one for its flat neck, which would allow me to design/make a simple seal. The goal is use a 350mm glass globe eventually, that will of course not fail under vacuum.
- the globe base a bunch of fittings for gas fill up and vacuum (I can bring the pressure down by 0.1-0.15 bar safely but that's enough to see a significant improvement on the arcs/plasma).
- 1.5" Brass ball


What's next:

- Short term: rewind yet another secondary, this time I'll add one layer of kapton tape to stick each new wire coming in a new chamber against that separation and keep it there, all the way down. I'll wind a bit slower too, to ensure I layer those turns as evenly as possible. I'll also try to not wind all the way to the top like I did on my first attempt which was overall much better considering it failed at much higher voltage and with a failure mode that can be completely avoided by skipping that last chamber (like I did on the second attempt).
- I am making an improved PTFE core, with thicker ends, and thicker body around the ferrite.
- I am redesigning everything else as well, larger diameter container but also much shorter. I'll also get the HV ground to come down from the bottom of the container rather than the top. I am designing this around a much bigger 20L boiling flask (that was the initial idea), and will be using vacuum valves (2 of them) for fill up and vacuum. I am also going to be using a large test tube (25mm diameter, only those are long enough to reach the center of the flask) as HV electrode support. All of those 3 tubes will go through the rubber cap of the boiling flask.
- I'll probably stick with my home made ZVS driver, unless I can figure out a way to use my SSTC driver (Full bridge, TO247 IGBT's). But that would only possible if I can get anywhere close to soft switching. And if I do stick with the ZVS driver, I need to research more what David mentioned about ZVS driver working better with gapped transformer which is not my case.
- I'd also like to add some kind of interrupter to the system (which would be trivial if I used the SSTC driver).

Pictures:

- Secondary Rev1.0 before failure
- Secondary Rev1.0 after failure: the black area has a pinhole pierced by the arc that went from the HV'most enamel wire through the PTFE to the ferrite
- Secondary Rev1.1: freshly wound secondary, this time with the last chamber skipped. Notice how, out of greed I added more turns to compensate for one less chamber? haha
- Driver pictures: I CNC'ed single sided Clad (4 or 5 oz I think). I used SOT227 Mosfets mainly for practical reasons; also thought I'd easier to kill them, especially considering I'm using current limited PSU's. I need to upgrade the 470 ohm resistor to higher wattage.

I'll post some stuff soon in a thread about the DR that I am putting together reusing this full bridge initially.

Meanwhile, I wanted to do an experiment with this bridge and driver. I didn't want to create a new thread for this especially as it's directly related to the stuff built in this thread. Basically I turned a multi-chamber secondary structure out of PTFE and got around to wind it today. I initially hooked it up to one of those ZVS boards with underwhelming results. It's a possibility that my primary didn't have enough turn (I went with 2x 2.5 turns - the driver is designed for a center tap primary). Anyway, I thought I'd give a shot on my SSTC driver/bridge. So I disconnected the 7414 and feedback, and taped my signal generator to the input pin of the driver chip. Results were quite a bit better than with the ZVS driver. And since I still had the oscilloscope hooked up the same way it was for the SSTC I immediately saw the IGBT were hard switching. With very little input power my temperature probe already showed they were heating more than they do with the SSTC at much great input power. I tried adding the DC blocking cap again thinking it might move the phase a bit, and it did but nothing significant. First screenshot is without the 4.7uF and the second is with it.

Anyway I stopped there and started doing some research on the topic and found that Steve Ward actually did something very similar: https://www.stevehv.4hv.org/FBD.htm
He doesn't talk about hard/soft switching but other than for a half bridge and a couple of caps it's basically the same concept. He does mention he was able to pull 450W which is probably over 10 times what I was able to pull safely today. IGBT's may not be the best for this, but I was comfortably switching at 150 kHz which is well in the capabilities of those IGBT's. With this soft switching, this project could work. The transformer itself (well the secondary) is a success from what I can tell. I have at least over 3000 turns, everything submerged in mineral oil.

I kept the same wire spacing, so coupling definitely went down as well.

Thanks for the corrections on the current calculations. I used the formula both Kaizer and Loneoceans shared on their SSTC pages but didn't realize they didn't have a full bridge as example. That makes sense though, I kinda figured that out by thinking about how the full positive bus generates the positive current waveform, i.e. ~340V for 0-Max current, and vice verse for the other half.

That final plot was in the part of the waveform where the amplitude is pretty much constant (i.e. a bit past the blob). That was more of a capture to get a measurement of the period to better approximate the reactance off the measured Fres rather than an estimation from JavaTC. I did gather a lot more data and didn't want to abuse of the great help and attention you've been giving  this thread; I also noticed how that one plot shows some deviation from the usually much cleaner sinewaves. This, (and the phase shift) happens relatively brutally around 180-200VDC on the bus, i.e. pretty clean before that. That also somewhat coincides with the phase shift, where under 180VDC current would switch close to 0 current, and it gets increasingly de-phased at higher voltages. A video would have probably been better to get more of a dynamic feel on how that shift and sine deformations are taking place.

The 20 ohm + 1 uF in series of the existing 1 uF DC blocking cap on the GDT didn't help, in fact it looks like the waveform hasn't changed, and the low frequency ring is still there.

I've made a short video which will show a bit more how that current waveform evolves from 0VAC to 110VAC: https://www.youtube.com/watch?v=_LIvVoRE4zg&feature=youtu.be

here's the current state of performance: https://www.youtube.com/watch?v=GMmQSiZCmdc&feature=youtu.be

I may be a bit too cautious, but I'm trying not to kill the coil or at least maximize time between repairs so I can gather as much data as possible to hopefully make sense of the physics behind all this. But I think my 1% duty cycle is just too little, that is for the tests I ran today. I am trying to not push more than 1ms pulses (which is what I deducted as a safe pulse duration from the IGBT datasheet). And since experimentally 10 Hz, 1% (1ms) gives me the longest arcs (rather, the best ratio of arc length over power consumed), they became my go-to tests parameters.

My secondary system is probably a little bit on the high side in terms of impedance (60kHz). By the way, running longer pulses is something I've tried for quick tests and they do help a lot, making arcs thicker and deeper. They appear longer but not sure if they actually are or if they just appear that way being so much brighter.

I've put the current probe on the secondary a few days ago just to have a look, I'll do that again and correlate that to primary voltage. I do wonder if this phasing is dependent on spark load.

One thing that has been consistent throughout all the tests so far is the shape of the current pulses. Aside from that low frequency ring, regardless of the enable parameters (frequency/on time) I always get that "blob" that's usually right in the middle of the low frequency ring. The shape of the blob will vary slightly based on the bus voltage (and as seen today, by the primary turns). What is causing the current waveform to have a stronger P2P in that early region of the pulse? I understand why it takes some time for the current waveform to get a certain regime (i.e. the feedback loop to lock on the resonance frequency), but then why does it drop/stabilize lower? Spark load will affect the resonant frequency but the feedback is there for that. And it can't possibly be capacitors discharging considering the energy stored and the fact that it does stabilize for the rest of the pulse, a cap discharging wouldn't - the time scale of those pulses (1ms) is a fraction of the mains' period.



Edit: I calculated an inductance of 24 uH for the primary as it currently is now. At 117 kHz (measured from screen 90) it should yield a reactance of 17.5 ohm, which at 320VDC should mean a peak to peak current of 18.5A or so. I measured (screen94) ~40A peak to peak away from the initial blob... I'll check my probe - seems like a big discrepancy if we consider the 40A - but on that screenshot the blob peaked over double that value. That same primary, with the 4.7uF DC blocking cap has a resonant frequency of just under 15 kHz, which incidentally still matches the low frequency ring. I'll have to measure it more accurately. Maybe I'll try a more tightly coupled primary again. You said the GDT might resonate with the driver's DC blocking cap, how would that cause the primary current to react that way, the primary circuit would need to resonate with that as well, wouldn't it?

Yes, the gates traces aren't as steep as they were with the half bridge. It won't hurt, but perhaps affects performance a little bit. I spent some time gathering data for various numbers of primary turns. Nothing really conclusive, I'll post about it later after I look at the numbers in more details.



By the way I didn't mention but that nothing on the bus screen capture triggered off the end of a pulse (i.e. when the enable pin of the driver goes low). It does oscillate for a while.

As opposed to the same capture but this time captured off a rising pulse (when the enable pin goes high, see attached screenshot where we can see both oscillations at the beginning of the pulse and at the end). Weirdly enough, it doesn't oscillate for nearly as long.

Tomorrow, I'll add the RC damping you mentioned.

edit: I just remembered, I had the same waveform on the scope (when nothing was on the Vbus) with the half bridge and more importantly with the previous GDT. Although, it's not impossible that both GDT's would resonate with the DC blocking cap at roughly the same frequency, how likely would that be though?

Also, looking how that low frequency ring affects the shape of the current waveform, how likely is it that it affects performance?