High Voltage Forum

High voltage => Transformer (Ferrite Core) => Topic started by: Andrew321 on January 12, 2020, 05:36:46 PM

Title: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 12, 2020, 05:36:46 PM
I've been playing around with Mazilli ZVS drivers and HV flyback transformers for about a year now and I decided to make a few alterations that just aren't panning out. My first driver was the most common one that you see online. I cant seem to figure out how to attack pictures, so this link will have to do.


The only adjustment I made was I used the IRFP260N. It worked perfectly. I got a few sets so I could play around with stuff and not worry too much about if I blow a MOSFET out. I dont have any testing equipment other than a DDM so a lot of my learning is trial and error.

My set up is simple. I have a largish variac which is rectified and hooked up to a large capacitor bank. I built my own flyback transformer because I ran out of ones salvaged form old TV's. This way if something burns out, I can just rewire it. Its made of 2 cores glued back to back with the primary in the middle and two HV secondary's on the outside connected in series and has a core gap of about 2mm, submerged in mineral oil.

With the IRFP260N I have no issues at all, but it tops out about 60V input and I wanted higher input voltage. The alterations I made were that I use 6000 ohm 6W resistors for the gates and I used a 24v zener diode and the IXFX100N65X2 MOSFET. When I look at the ratings of the IXFX100N65X2 vs the IRFP260N it seems like it should work, but it doesn't. I got 6 of them (trial and error) and all of them have failed as soon as I turn them on and I cant for the life of my figure out why. The only time it doesn fail instantly is when I turn them on at about 20V and I can hear the transformer ringing but as soon as adjust the voltage or try to draw an arc it fails. They don't blow up, its just a sudden dead short in one of the MOSFET's. I'm determined to make it work, however I'm a bit out of my depth and until I know why its failing I don't see any point getting more or trying new MOSFET's. Especially at $20 a pop.

I did some reading online and it seems that running them at higher voltages is a problem because of, and correct me if I'm wrong, the output capacitance of the MOSFET increases with the voltage rating and that doesn't allow the body diode to recover fast enough? Either that or something to do with bipolar latchup (which I still need to read up on). The one article I read says that reducing the switching speed should help, but I'm operating at levels I can hear, so under 20kHz, and that should give the diode plenty of time to recover.

Can any one shed any light on this for me? Any advice or linke for reading would me much appreciated. I salvaged a pair of the MOSFET's for one more attempt, but don't want to make a move until I get some solid input.

here are some links to the parts and the most helpful article so far:



Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 13, 2020, 01:24:36 AM
It will be an interesting challenge to debug with only a DMM for measurement.  However, a couple thoughts come to mind.

First, at such low frequencies and relatively high voltage, saturation of either the supply inductor or the flyback core seems likely.  Can you share specifications or design details for those?  For the flyback, how many turns (especially for the primary)?  What core material and dimensions (especially cross-sectional area within winding)?  What core and winding count is your power feed inductor?  If a commercial inductor, current and inductance specifications?  What resonant capacitor(s) are you using - capacitance, voltage, dielectric material?

Second issue is gate-drive current.  6k ohms on 60V is 10mA.  That will take 18us to charge the 180nC gate charge, too slow for even your low-frequency operation.  For high-voltage operation, I'd suggest the following ZVS circuit.  It's what I use in my ~1kW 170V Jacob's ladder (documented under it's own topic):
 [ Invalid Attachment ]

Another circuit option for lower gate-drive impedance is:
 [ Invalid Attachment ]
This is closer to the typical circuit, but uses two small PFETs to turn off the gate-drive resistor current during most of the cycle, allowing much lower-ohm resistors without excess power dissipation.  I prefer the first option above, but either should work for your needs.  This second one is designed for 60V, but values could be adjusted for higher voltage.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 14, 2020, 02:27:54 AM
I dont know how to calculate the time it takes to charge the gate (if you could share that would be much appreciated) but I suspected that 6k ohms was too high. So I went ahead reduced the gate resistor to 1k ohm and turned it on at 25VDC. It worked perfectly and I could draw an arc. I cranked it up to 100VDC and it was still working, but at max arc draw the MOSFET's shorted and were destroyed. So I think we are on the right path here.

The prints you provided were very helpful and I'm still reading though your post on the Jacob's ladder you made. Interesting stuff and very impressive! I have a few questions about the Jacob's ladder circuit, but I'll get to asking them after I read those few pages on it. I'll most likely be ordering parts to make one.

To answer your questions:

The supply inductor (I'm assuming you mean the toroid) is one I got form a discount store. Its ID is 23.5mm, OD is 47mm, height is 18mm and its blue. I wound 17 turns of 14guage wire on it.

The transformer cores I got on amazon. They are UF86A,U shape, MnZn PC40. I got two pairs and glued them together so it looks like 2 E sections where the center is thicker than the outsides. I forget exactly, but I believe I put a total of 14 turns around the center part of the E and center taped it (made some litz wire for it). The cross sectional area of the primary coil is about 855mm (its 28.5mm x 30mm) and has a 2mm gap between the two E sections. I put too many coils on the secondarys to keep track of, but should have about the same amount.


The capacitors are MKP, 275nF, 530VAC. I use 3 of them in parallel for a total of 824nF.

I guess what I'm most confused about is why it works with the IRFP260N and  sort of not with the IXFX100N65X2. The IRFP260N circuit has a 2.5k ohm gate resistor and a Qg of 234 nC vs the IXFX100N65X2 Qg of 184nC.

I know its next to impossible to figure out what's going on with only a DMM. I'm thinking about getting something to measure inductance and a cheep scope. They have pocket size scopes on amazon for about $80. Should serve my needs.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 14, 2020, 05:27:11 AM

The supply inductor is entirely a guess without knowing core parameters.  It's presumably powdered iron or iron alloy.  Taking a wild guess at permeability of 100, it would be about 70uH and good for ~70A peak.  The flyback is easier to estimate, as PC40 material characteristics are published.  Still guessing on the fringing fields around the 2mm gaps.  I'm estimating 60-70uH for 14 turns, but that would make the frequency a bit over 20kHz.  Hearing the oscillation suggests the frequency may be lower and therefore the inductance higher.  Perhaps 16 turns or perhaps more fringing field than I'm guessing.  Either way, it should handle ~700V peak, or ~200V DC input before saturating.  So it looks like saturation is probably not the issue.

That leaves two likely causes.  Slow FET turn-on is one.  The two FET gate charges are quite similar.  The primary difference is the gate-source charge (charge to the first knee in gate-charge graphs).  It's only 25nC for IRFP260N, but 60nC for IXFX100N65X2.  The second knee is more similar, 100nC vs. 120nC.  These curves are not directly relevant to ZVS use.  Simulation would be the best way to analyze the differences.  There may be some artifacts when the FET does belatedly turn on - I haven't simulated or experimented enough to know what issues may arise.  (LTSpice is a free simulation tool that I use regularly.  It can be downloaded from Analog Devices, since they bought Linear Technology.)

The other possible issue that I didn't think of initially is the arc causing enough load to make the resonant Q too low to maintain oscillation.  When ZVS oscillators drop out of oscillation, both FETs are on with ~6V drop, and the current rises to whatever the power supply can feed to a 6V load, often enough to fry FETs.  (My Jacob's ladder has an over-current shut-down circuit that I didn't detail in the post.)

In both simulations and experiments, having a coupling factor of 0.85 or lower allows the ZVS to continue running with any secondary (arc) load.  For low-impedance loads (short circuit), the oscillation frequency is ~1.5-1.8x higher, being defined by the leakage inductance.  With no arc (high-impedance load), the frequency is normal, based on the flyback transformer's inductance.  I tried winding an old TV flyback with 10 turns on the exposed U-core leg and measured coupling factor at 0.81.  That's the typical configuration for the instructables.  I'm guessing your coupling factor may be low enough, but can't tell well enough from just geometry.  Drawing an arc will increase the resonant frequency, so make FET switching time more critical.

One other complication with HV transformers is that the HV winding often has enough capacitance to have a major effect on the resonant frequency.  This capacitance is after the coupling factor (leakage inductance), so behavior gets a bit complex.  I'm not too sure how well my <=0.85 coupling factor rule holds when there's significant secondary capacitance.  (I'm currently making a small ZVS HV circuit where the secondary winding capacitance is the dominant resonate capacitance, with only tiny capacitors on the primary.)  Perhaps secondary winding capacitance explains why the frequency is low enough to hear, rather than 16 turns as I had guessed initially.

I'd suggest getting the switching times well under 1us for charging to the second knee (120nC), which implies a current over 120mA.  That's the advantage of my Jacob's ladder circuit, the gate-charging current comes from the resonant circuit directly, rather than from a gate pull-up resistor.  Once gate drive is fast, it will be easier to hunt for other issues.

Just remembered one more possible issue to warn about, especially if you continue past 100V.  Sudden load changes can cause momentary spikes in oscillation voltage.  As the arc length grows, power increases, so the current in the power-feed inductor increases.  When the arc breaks, power suddenly drops.  The power-feed inductor has lots of stored energy (I^2L).  This stored energy feeds into the primary ZVS oscillation, which increases amplitude until all that energy is stored in the resonant circuit (capacitor and flyback inductance).  This can more than double the oscillation amplitude, so could even be a problem at 100V.  100VDC input is ideally PI times that or 314V peak.  Not much more than double that will reach avalanche-breakdown Vds for the IXFX100N65X2 FETs.  I use lots of 1500W TVS diodes across the IGBTs in my Jacob's ladder, as it has a very tight voltage budget.  170V peak input (120VAC rectified line) leads to 535V normal peak, not much below the 600V IGBT rating I'm using.

This was a rambling answer, but hopefully useful.  Good luck!
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: T3sl4co1l on January 14, 2020, 07:36:14 PM
FYI, this type of circuit doesn't much care for switching times -- that's the whole point and meaning of ZVS -- but do check if the slew rate is adequate, i.e. because of the resistor pullups there.

The one thing you definitely do not want to do is enforce sharp switching.  This circuit is an analog oscillator and anything you do to break that loop, will break the oscillator.  It may start up in a random oscillation mode, it may sit there chattering and burn up transistors even faster than before.  Without logic, it may even turn off both transistors at the same time, which in a current-fed inverter is precisely the same error as turning ON both transistors in a half-bridge!  Kaboom!

Also note that the opposing side may not pull down far enough to turn off the one side, i.e. you need low drain saturation voltage.  And this includes dynamic voltage, so ground impedance needs to be quite low, as well as Rds(on).

Which makes high voltage operation that much worse, and in turn, high power operation that much worse.  This circuit works best, say, under 100V?  The problem is, drain and gate voltages are connected (by the diodes).  Drain voltage being comparable, makes sense (indeed, it has to be more than ~5V, otherwise the gate won't fully turn on).  Drain voltage being more than, say, 10 times higher -- not as sensible, it puts heavier demands on the transistors.  Still possible, yes, but it just keeps getting less and less reasonable as it goes up.

Which means even if you're dumping upwards of 50A, you're limited to a few kW, maybe 10kW tops, and you'd need a bunch of transistors in parallel to reach that level I would think.  And you still need a power supply or converter of some sort, it's not direct offline range.  So you're dealing with those knock-on costs as well.

Last time I build one of these circuits, I used a complementary emitter follower to buffer the gate voltage, so a modest (1k?) pullup was all that's needed, and that ran at 500kHz into transistors, that are, smaller than what you have here (~30nC??) but because the frequency is so much higher in my case, it illustrates much higher performance.  So do consider using that.

The second circuit above is interesting in that the gate pullup is switched, allowing reasonable performance without costing too much continuous dissipation.  Probably performs as well as my configuration, give or take optimization of component values.  Probably a bit less efficient overall (you're still dissipating some excess power in those resistors) but not nearly enough to matter for us.

For high power, I would rather suggest a half-bridge circuit, which can be very much like the current mode circuit in a CFL.  Now, it is current mode, which stinks for MOSFETs which are voltage mode drive.  You could perhaps assemble a little transresistance amplifier (current in, voltage out) to convert current transformer feedback into gate drive voltage, without sacrificing the analog oscillator behavior; which again, this circuit necessarily has.  The half-bridge configuration allows a voltage-fed inverter, and divides the supply voltage so that high voltages are economical to use (500V devices with rectified 240VAC, say).

Note the load will be series resonant, and should be a relatively high impedance, so, a smaller capacitor (rated much higher voltage) and more primary turns.

One advantage will be higher OCV, easier for striking sparks.  Actually, it may be too high, which will also draw excessive current from the inverter, and drop excessive voltage across the capacitor.  You may want an overcurrent circuit.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 20, 2020, 05:55:07 AM

I'm curious what your experiments will show with the secondary capacitance being the main resonating capacitance. Will be looking for that in the future.

I had to do some googling on leakage inductance and coupling factors to start to understand what you were talking but, but I found it really help full. I think your last point about the high load dropping out and causing a voltage spike is spot on. When I got the new one running (in my previous post), it was working fine until I went a bit over 100VDC input and then at the end of the arc the MOSFET's were destroyed. So I'm thinking I will be adopting second circuit but will add some TVS diodes. I do have a few, maybe amateurish, questions for you about it.

First: What is the purpose of C3?
Second: What are the important characteristic of the p-mosfet that I should be keeping an eye out for?
Third: When selecting the TVS diode, I would want to choose something a few tens of volts under my MOSFET voltage rating, but also well over normal operating voltage, correct? Why did you chose such high wattage TVS diodes?
Lastly: Would you mind telling me if something like this MOSFET is a good fit? I looked for something with a smaller gate charge and slightly higher voltage rating, as that seem to be my main issue.


Also, I had a bit of a face palm moment. I realized that the reason for the unexplained failure of the brand new MOSFET's could be due to them being about a foot away from my X8 voltage multiplier with an estimated output of 200'000 volts. I noticed while running it today that some metal objects around it started arcing to ground. Probably what did them in. The ones that did work I think survived by luck and died due to the voltage spike caused by the sudden drop in load.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: SteveN87 on January 20, 2020, 12:59:19 PM
What is the purpose of C3?

That looks like (along with R3) a soft start circuit. I used a similar scheme with a flyback driver with very low primary inductance; without it, the power supply was tripping before the circuit could start oscillating.

(Apologies it that's completely wrong!)
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 20, 2020, 09:48:06 PM

Just got my new little (20W) ZVS running yesterday, the one with stray HV winding capacitance of the transformer as the dominant resonance.  Matches simulation fairly well.  Normal intended oscillation is 22kHz - feeds a small DIY plasma ball with ~+-8kV.  As with Tesla coils, it has a second higher resonant frequency where the transformer primary and secondary winding currents are 180 degrees out-of-phase, 530kHz in my case.  As in simulation, I had to add a 530kHz notch filter across the primary to prevent locking to that frequency instead of 22kHz.

C3 and R3 form a local 12V gate-drive DC supply.  For that version running on only 19V input, picking a value for R3 got me close enough to 12V.  For higher input voltages, place a 12V zener diode across C3 to keep the gate voltage at 12V.  (I added a 12V zener for my 60V version of that circuit.)  Unlike the typical ZVS circuits, the average current needed by the gate-drive 12V supply is fairly low.  However, the current draw peaks at each zero-crossing.  C3 is just a supply bypass capacitor to provide those momentary current peaks.  You may need to adjust the value of R3 to provide enough current to reach 12V, but not too much excess to avoid wasting power in R3 and the zener diode.  Or, if it's easier, just use a small 12V DC supply (wall wart or ...) for the gate drive voltage.  Keep C3 for local current peaks.  Value is not critical - I had a 10uF cap around, which measured 9uF, so that's what is in my schematic.

Startup of ZVS oscillators can be problematic.  In the first circuit I shared (Jacob's ladder), R2 provides a lower current to the main oscillator to get it started before applying full power through switch S1.  For the second circuit, I don't recall if C3's slow charging helped with startup or not.  At only 19V where I simulated, it worked as is.  I'll need to play a bit with higher voltages in simulation.

The small PMOS FETs see only the gate-drive voltage, so 20 or 30V parts are fine.  Their gate voltage is the same as for the large devices.  That's typically the issue with such low-voltage devices - they often have 8 or 12V maximum Vgs.  I have lots of old (obsolete) small 60V FETs (VP1306) around, so use those.  Just look for a small PFET with on-resistance under 10 ohms and max Vgs of 20V or more.

TVS diodes are listed by the peak transient power they can handle.  1.5kW TVS diodes are in similar packages as 3-6A rectifier diodes, and are typically rated for 5W continuous power.  Voltage specifications are a bit complex too:  Standoff voltage, breakdown voltage, clamping voltage.  For hobby projects, I measure low-current breakdown voltage after buying a set, then use them at that voltage.  For example, 100 1.5KE250CA parts measure very close to 250V at room temperature and ~1mA current.  Pairs of those in series start conducting at 500V.  Even though my Jacob's ladder theoretically hits 530V peak, the TVS diodes just warm up a bit until their breakdown voltage increases enough to handle that.

I don't have any personal experience with SiC parts, but the UF3C065040K3S one you listed looks good.  Definitely lower gate charge.
Do you know how much current you will be running from your power source?  One issue with higher-voltage ZVS using FETs:  As the FETs warm up, on-resistance (and therefore Vds) goes up and gate threshold voltage goes down.  With the diode-drop added, the FET in the ON state may not have low enough Vds to keep the other FET turned-off completely.  (That's one small advantage to my first circuit.  It avoids the diode drop, so drives the OFF-state FET with slightly lower Vgs.)  Have you considered using IGBTs for your relatively low-frequency ZVS?  IGBTs have lower temperature coefficients of threshold voltage and especially forward voltage drop (Vce).  That's why I use them for my Jacob's ladder circuit.  Even at it's ~100kHz frequency, there are fast enough IGBTs available.

Proximity to 200kV could be an issue if your circuit isn't shielded.  If it is in a shielded box, it should be fine.  (Not being mechanically inclined, my circuit shielding is often a cardboard box lined with aluminum foil.)

Good luck!
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: T3sl4co1l on January 20, 2020, 10:55:35 PM
TVS diodes are used this way:

Nominal rating: maximum operating voltage while drawing small currents, i.e. Ir <= Ir(max).
Breakdown: voltage range (mfg spread) at specified Ir = Iz (usually 1mA or something).
Peak: voltage max at specified Ir = Ipk (usually some amperes).

Ipk is the permissible peak current for the given waveform, usually a 8/20us (direct or induced surge) or 10/1000us (induced lightning) transient.

Roughly speaking, Vpk is Vbr + Rs * Ipk, assuming a resistive characteristic.  The avalanche breakdown mechanism is pretty aggressive even at these high currents, so it's still not resistive as such.

Typically, Vpk is around 30% higher than Vnom, so you want to choose Vnom for the application, then Vds(max) (since we're talking about protecting transistors here) to be >= 30% higher than that.  650V transistors should operate up to about 400-500V peak, and use a TVS (or stack thereof) rated the same nominal.  Which, for a PP ZVS circuit, is a supply around 220VDC.

Since Vpk depends on Ipk, if you know the Ipk in your application, you can solve for a more representative Vpk.  The datasheets aren't usually specific enough to do this, however.  (MOVs do usually give these data, but MOVs are much sloppier than TVSs -- higher internal resistance, bigger ratio between Vnom and Vpk -- so they aren't so great at protecting transistors.)

Because Ipk is defined by a given waveform, if you know your waveform is different, the maximum permissible Ipk may be different in your application as well.  TVS have pretty good dynamic range, good enough that you can reasonably extrapolate the power-time curves down to quite small time scales (fractional microseconds?).  Which usually follows about a P ~ 1/sqrt(t) asymptote, implying the limiting factor is heat diffusing out of the junction.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 22, 2020, 05:45:43 AM

Oh man, every time I think I got my head around how these little magic black boxes work I learn some new parameter I need to keep in mind lol. Normally I just look for highest Vds, lowest Rds on, 20-100amp continuous, lowest time delay on/off and lowest rise/fall times. I just added lower gate charges and faster revers recovery times. Now I gotta look into how thermal co efficient works. Deviating from the tried and true comes with quite the learning curve! I don't mind it though.
 I thought about using IGBT's but I know less about them and their requirements than MOSFET's and so have been hesitant to use them. But now that I am looking at them.... Seems like I should be using them. For $6.59 a piece, what's wrong with this?

I cant pull more than 15 amps from my home plugs, so anything more than 20 amps should be fine. The one above is 80amps and 1200v for Vce. That means I wouldn't have to use TVS diodes. I kind of feel like a sucker for paying $20 a pop for my last set of MOSFET's lol.

When you say DIY plasma ball, do you mean light and incandescent bulb? I have a plasma ball at home and as soon and I learned I could pop the ball off and put it on my ZVS flyback I did. Less than spectacular and I think its because the frequency was too low, probably around 10kHz. When you did yours, did you connect the other ent of the HV secondary to ground?
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 22, 2020, 05:54:17 AM

I don't think I'll be using the H-bridge because that would mean I need ot learn a whole new thing. I feel like im so close to making this work!
My whole goal is to get an out put of 100Kv on the secondary and I think its doable. I took some measurements the other day and about ever 5v into the ZVS driver gets me about 2500v on the secondary. That's a low figure because the resistor I'm using as a probe for me DMM draws power plus I'm losing voltage due to harp points in open air. At 60VDC in I get about 30KV out of the secondary. I figure if I can pull this off, at 170VDC in I should get some where between 80KV to 100KV on the secondary. From there its a x10 multiplier and I've got 1MV, which has been my goal for ages lol. But if I can get this ZVS driver up and running ill take more accurate measurements in oil with larger capacitors.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 23, 2020, 05:57:18 AM

Yes, there are a lot of parameters, and it can be difficult to figure out exactly how they affect specific circuits.  For added difficulty, IGBT simulation models are relatively rare.  It's harder to accurately simulate their detailed behavior compared to FETs.  However, their cost for a given current and voltage capability makes IGBTs attractive.  I use them much more now.

Despite what Tim said, switching speed does matter at least some.  In normal operation, the FETs or IGBTs conduct just the current flowing through the power feed inductor, which is the power input current + inductor current ripple.  During switching, while waiting for the conducting IGBT to turn off, the IGBTs conduct the resonant current, which can be much higher than the supply input current.  The voltage across the resonant capacitor momentarily stops changing, so all the resonant inductor's current (transformer primary current) now flows through the IGBTs, in the forward direction for the one turning off, and through the internal anti-parallel diode for the one about to be turned on.  To minimize issues, I'd recommend a copper ground plane connecting the two IGBT emitter terminals, short emitter lead length, and short wires or copper planes from the collectors to the resonant caps (your three 275nF caps).

The part you selected should be fine.  A couple faster (and a bit cheaper) options:

Notice that IGBTs are generally spec'ed at 15Vge, so use 15 to 20V for the gate supply rather than 10-12V.

15A RMS from the line is 22A peak.  Including start-up spikes and inductor ripple, I'd suggest IGBTs capable of pulsed collector current of 60A or more (which your part and the other two I suggested all meet), including low enough Vce at that current to keep the other IGBT off.

BTW, I've now posted my DIY plasma ball:
Yes, the HV transformer winding is grounded on one end and drives the center plasma-ball electrode from the other end.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 25, 2020, 02:22:09 AM
The parts have been ordered! They should arrive some time next week and then I'll put the thing together and let you know how it goes. If all is well I might even put pics or a video :)

Talking about the switching times, I have a question for you. What exactly is it that limits the switching speed of the device? Is it the t on/off and rise/fall times and the reverse recover time? I normally aim for under 100ns for any one of those when I look around. The IRFP260n for example has a reverse recover of 402ns and adding up all the time on/off and rise/fall times gets me 180ns. Combine those two values to get 582ns. Then double it. Why? Because I don't fully understand how this stuff works and I want to be on the safe side. Also because I figure a sine wave will change directions twice in one wave length. That brings us to 1164ns. Convert that to frequency is about a theoretical max of 860kHz? How far off am I with all this?
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: T3sl4co1l on January 25, 2020, 06:37:47 AM
Reverse recovery is a diode thing, not a MOS thing.

If you use two transistors in a synchronous buck or boost converter, for example, you will be waiting approximately t_rr when the high side turns on again, for commutation to occur.  (In the mean time, switch current rises to, well, about I_L + I_rr.  It's pretty nasty.  The reason why high speed PN diodes have been improved so much over the years, and even schottky diodes surpassing them in turn for most applications.)

Indeed, it can be better to leave no deadtime between device switching in this case, so as not to give the body diode an opportunity to forward-bias.  The MOS turns off ~instantly, whereas the diode "drools", relatively speaking, for ages.  Mind, this isn't always possible to do, or practical.  Controllers and drivers rarely provide adjustable dead time; manufacture, temperature and aging variations, may cause the dead time to drift +/-, so that your golden prototype might be perfect, but maybe the production yield is awful.

Anyway, a ZVS circuit undergoes recovery early in the cycle (if the diode conducts at all), not late, so it's never a concern there.

And again, in a self-oscillating circuit, fast rise times aren't important or necessary, or actually even desirable.  You can't have infinite switching time, of course, it needs to work in the first place; but whole microseconds is quite slow switching indeed compared to digital circuits, but is perfectly acceptable here.

Which is another reason why resonant circuits can operate at higher frequencies.

In digital circuits, switching time is determined by drive strength (driver resistance or current), internal gate (spreading) resistance, stray inductance (especially source inductance that is shared between gate and drain paths -- Kelvin connections can avoid this, hence packages like SOT-227, TO-247-4, D2PAK-7, some DFNs, etc.), and the capacitance curves of the device.

Note that drain voltage swing can be quite rapid -- important for keeping switching losses low -- even while gate voltage is changing relatively slowly, when an inductive load is present.  This is especially true of newer generation parts, which have a very sharp Crss(Vds) function.  At high Vds, Crss is small, and so there is little Miller effect; as Vds falls, Crss increases several orders of magnitude, forcing Vgs to slow down -- commonly called the Miller plateau.  After Vds falls to nearly zero (saturation), Vgs continues rising, onward to the supplied Vgs(on).

Switching loss can be estimated by the triangular area created between on and off conditions, as Vds and Id swap from supply voltage and zero amps, to zero volts and full load current (and vice versa).

For an inductive load, on a voltage-sourced bridge, current lags voltage; so, the voltage swings first, and then current drops.  Commutation and loss occurs on the turn-off edge, while the turn-on edge is ZVS.

For a linear (constant capacitance) device with constant-current gate drive, at first nothing happens (Vgs = Vgs(on), then the voltage ramps up (constant dV/dt), then the current ramps down.  The current doesn't begin dropping until the load current is caught by the opposite side diode or switch.

If we draw V, I and V*I on a graph, we find V*I is a triangle of width t_sw and height (peak power) Vsupply*Iload, thus its area is half the product, or: Esw = 0.5*Vsupply*Iload*t_sw.  The average switching loss is then this times Fsw.  (Note that I'm using t_sw for commutation time, but Fsw for switching frequency...)  t_sw is the time taken from Vgs(plateau) to Vgs(th), not the full gate voltage edge -- at Vgs > Vgs(plateau), the MOSFET just gets slightly better saturated, while at low voltages it just gets more "off", so there is "wasted" gate voltage swing, with commutation only occurring in the middle.  Consequently, Vds rise/fall is always faster than Vgs fall/rise.  Typically by 2-3x, so you can get a rough idea what Vds should look like based on gate swing.  And you calculate gate swing from Qg(tot) and driver+gate resistance.

Because real MOSFETs have nonlinear capacitance (C varies with Vds, or sometimes Vgs or Vgd), the Vds edge can be much faster still, which reduces power dissipation.  This is hard to calculate from any given set of parameters, because no one parameter can account for the fact that the capacitance follows a curve.  You can use effective capacitance values, but which one is important depends on what it's effectively modeling (e.g., speed under a given load condition, energy), and the figure depends on the Vsupply it was measured at.

The upside is, switching is much faster nowadays than it used to be, which is good for losses but bad for layout.  And we can use much bigger devices (lower Rds(on)), making conduction losses lower, too.  And Qg(tot) is reduced so much that we're still spending less gate drive power in the process.

The downside is, all that extra drain capacitance at low voltages, can act very much in the same way as diode recovery does.  Even when you're doing a synchronous converter with the interleaved gate drive timing trick.

By the way, Miller effect is only driven by voltages, so the Miller plateau only accompanies the voltage swing.  Current swing does have the same effect, via source inductance.  The difference is, because the inductance is due to strays, you can't always measure Vgs as such, so you may not be able to tell this is going on!

Regarding switching frequency: because instantaneous switching losses are so high, you might only be able to handle 1/20th of a cycle as switching loss, if even that.  Driving an IRFP260 at less than 100ns implies fractional MHz switching, which seems reasonable.

I've seen, not IRFP260s, I think they were 460s, used massively in parallel (like 32 up), at 450kHz, for 50kW per module.  That was a... weird induction heater.  It was a current-sourcing inverter, which means they used fuckoff massive inductors to deliver that constant current, controlled with a much slower (IGBT, or even SCR?) front end.  Consequently, instead of a series LC, or series-fed parallel-tuned (LLC), topology, they used a parallel-fed series-tuned (CCL) topology.  Which was a pain to tune: you can't tune or tap a capacitor this size, you can only bolt in a different combination of parts!  That was compounded by the very narrow tuning range of the control (think it was like 400-450kHz or something like that).  Which is easily spanned when a steel workpiece is heated and transforms from magnetic to nonmagnetic.  So it wasn't even suitable for a lot of applications...

I don't know that anyone's tried (or bothered) using IRFP260 for RF amps, anywhere into the SW band (1.8-30MHz).  An RF amp is basically all switching loss, if you will -- spending most or all its time in the lossy linear range.  The limiting factors (again, drive and internal gate resistance, inductance, and device capacitance) limit switching speed of these to somewhere in the 10s of ns, or to the low 10s of MHz for RF amps.

Higher voltage, and smaller die, types have certainly been used.  I've built a 4W 50MHz linear amp myself, using a pair of IRF510s, and others have used IRF540 and larger types under less general (tuned amplifier) conditions, and to much higher power levels (1kW say).

Interestingly, most power MOSFETs don't drop off sharply at high frequency, they just gradually get worse and worse, and their impedances start flipping shit.  I built a small amplifier with 2N7002, which illustrates this cleanly.  Up to about 30 or 50MHz, it's... not really a good amplifier, because it takes so much current to get moving at that frequency (100-200mA), which leaves very little supply voltage (a few volts) before the SOT-23 part burns up.  Again, dependent capacitances and all -- it would perform much better if it could survive 10 or 20V, and a more modest 20-50mA would suffice, but no one makes a version of this in a TO-220 or whatever, so that's out.  Anyway, at higher frequencies, the impedances slowly drop (I think due to feedback -- Crss), and the gain drops.  It's not that the device has a cutoff frequency (like BJTs do: fT), it has still more bandwidth to go -- but it's not really worth trying to use it.  Think it was something like 17dB gain (in a cascode) up to 10MHz, then a few dB off around 20-50MHz, then -3dB/oct beyond there -- an RC filter is -6dB/oct, so again, it's not a simple cutoff.  Up at, say, 200MHz, it's still got real gain, 8 or 10dB, and probably by reducing source and load impedances another 1-3dB could be obtained up there.  The fact that it's still got gain up there, implies an fT that increases with frequency (fT = 200MHz when measured at 20MHz; 600MHz when measured at 200, etc.) -- again, yet another way of saying it's not a dominant-pole cutoff, it's weirder than that.

And this behavior is more or less representative of mid generation devices, so, including HEXFETs like the present subject.

As far as I know, a lot of newer devices use fractal gate and source connects, so that there are ~equal resistances to every spot on the die, and thus the gain cutoff is much sharper and higher.  New devices have significant gain at 200-400MHz (and can easily deliver 10% bandwidth), enough that a poorly laid out circuit can behave as a tone-burst oscillator, burping out tens of volts -- real power, at VHF-UHF -- for a dozen or so cycles, in that time where the gate and drain voltages are transitioning!

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 26, 2020, 04:55:49 AM

As Tim said, the diode reverse-recovery time applies to only the anti-parallel diodes within the IGBTs, and doesn't affect ZVS operation.  In cases of slow IGBT switching, the diodes do conduct.  However, they come out of conduction (current reverses) when the IGBT is biased on.  There is little voltage across the diode, so plenty of time to recover without wasting power.

Concerning IGBT switching times, the turn off delay and fall time (Td(off) and Tf) are typically the slow (long) times.  These aren't too problematic, but do cause momentarily higher IGBT currents during that delay.  I don't have any IGBT simulation models here at home, but here's a FET ZVS simulation with added gate resistance to slow transitions:
 [ Invalid Attachment ]
 [ Invalid Attachment ]

As you can see in the plots, the IGBT current is higher during the delay.  They momentarily need to conduct the resonant current as well as the supply feed current.  The only issue is making sure this transient current is within the pulse-current rating of the IGBTs.

Turn-on delay is usually plenty short with IGBTs.  If not, it would cause issues as Tim pointed out in an earlier reply on this topic:
"The one thing you definitely do not want to do is enforce sharp switching.  This circuit is an analog oscillator and anything you do to break that loop, will break the oscillator.  It may start up in a random oscillation mode, it may sit there chattering and burn up transistors even faster than before.  Without logic, it may even turn off both transistors at the same time, which in a current-fed inverter is precisely the same error as turning ON both transistors in a half-bridge!  Kaboom!"

Turn-on delay would cause the above issue of "turn off both transistors at the same time", which is an issue.

Also from the above quote, sharp-switching can be problematic.  Some IGBTs have a long turn-off delay Td(off), but a short current fall time Tf.  That's the reason for my previous comments about using a copper plane to connect the emitters and short wires/planes from the collectors to the resonant caps.  Otherwise rapid turn-off will cause collector and emitter voltage spikes due to wiring inductance, which can then over-voltage the gates.

So, to finally answer your question about calculating speed:
1) Td(on) and Tr are typically plenty short.  If not, there will be collector voltage spikes when both IGBTs are off.  I don't have any good simulations or test to show exactly where this becomes a serious issue.
2) Presuming short Td(on) and Tr, the maximum frequency is controlled by Td(off) and Tf.  The resonant LC pauses during this delay period between each half-cycle (see the above simulation).  Consider an example where the LC circuit is tuned to 500kHz (1us per half-period), and where Td(off) + Tf is 500ns.  That 500ns is added after each half-period, so the period becomes 2 * (1us + 500ns) = 3us (333kHz).  The output waveform is no longer sinusoidal, having added odd harmonics.

Even with 1200V IGBTs, I'd recommend including TVS diodes.   In the above simulation I added them after the supply feed inductor.  At least in simulation it starts oscillating much more reliably and cleanly with the TVS diodes (D4, D5, and D6) included.  Also added D1 to clamp the gate voltage.

Hope it comes together well!  Once built, a picture of the construction (IGBTs to resonant caps in particular) would help in diagnosing possible issues.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: T3sl4co1l on January 27, 2020, 12:54:42 AM
A note about published timings -- these are often as much (or more?) a product of how they are measured, as a device characteristic.  In general, they're more for information than actual design use.

Consider two MOSFETs of equal Qg(tot), one logic-level type, one standard.  The logic level type will show somewhat faster turn-on times and significantly slower turn-off times.  Why?  The same charge has to be delivered to/from the gate terminal; the Miller plateau (where most of the switching is done) occurs at a lower voltage for the logic-level type.  When these are tested with same source/driver resistance (usually 4.7 ohms or something like that), the logic-level part is at a significant disadvantage -- less current is drawn during the plateau when it's at a lower voltage.

But that just means you can drive the logic-level part with some negative bias (if you happen to have a negative supply handy), and get symmetrical switching speeds.  Or drive it with a lower resistance source -- but be careful that even schottky diodes have significant voltage drop relative to a logic-level part, so you have less freedom to set rise and fall times independently.

For the IGBT, it might be tested at Vge(on) = 10 or 15V, and Vge(off) = 0V.  The threshold is below middle, 3-5V or so, so the switching speeds are asymmetrical.

It's quite common to drive IGBTs with negative bias, Vge(off) = -5 to -10V, in which case the switching will be symmetrical.

Modules tend to be rated at Vge(off) = -15V, for a full symmetrical swing; but beware that Qg(tot) may not be rated at the same condition.  This can make for a nasty surprise when designing the driver!

IGBTs work much like a MOSFET into a PNP follower.  Fast IGBTs have a low-hFE bipolar component, so a large fraction of load current flows through the MOS part, say 15-50%.  When the MOS turns off, the load current suddenly drops by as much, then the remaining bipolar current drops exponentially.  (This does actually act like diode recovery, internally.)

Slow IGBTs afford a lower voltage drop and higher current density, at the expense of slower speed.  This comes with a smaller MOS current fraction (maybe 2-5%?), higher hFE and slower turn-off (the bipolar current flow "drools" for some 100s ns, or even us for the higher voltage types).

Because current drops relatively slowly, IGBTs can dissipate a huge amount of turn-off energy, and are measured differently.  Whereas a MOSFET might measure its timings from e.g. 10% drain voltage to 10% drain current (inductive turn-off), the IGBT is measured to, say, 2% current.

Exact thresholds may vary by manufacturer, and I don't remember offhand if they usually list this in the datasheet, or if it's in a different document, or not disclosed at all...  Keep an eye out.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 31, 2020, 01:26:49 AM
Tim, Dave, Thanks for your replies about the switching times. Half of it was over my head but it gives me something to study up on.

So initially I was going to use the P-mosfets but I realized I may have got the wrong parts.... but now that I'm looking at the last circuit you provided, Dave, I think I may have just misunderstood. No matter, I'll try again tomorrow. But seeing as I thought I had the wrong parts I figured it couldn't hurt tiring it out the old way. So I hooked it up with each gate having two 220ohm 3W resistors in series (440ohm 6W total). It was working right up to 80VDC when I could smell something getting really hot. I quickly shut off power and the shoulder joins linking the two series resisters melted and they fell apart! So I reorganized what resistors I have to try and spread out the heat a bit more. But still, it got really hot at 100VDC input. So I tried to figure out the math behind how much power those things are dissipating. Correct me if I'm doing this wrong, but I figure:

80VDC / 440ohm = 0.182A x 80VDC = 14.5W!

That's huge! And if I were to get my 170VDC input it would be even more. I don't think I want resistors that big. So I got a small transformer I had laying around. After rectify and filter it I get 25VDC and after I reduce the gate resistors to 220ohm it delivers 0.1A. Probably good enough.

So the first attempt failed at 100VDC because my set up is messy and I fumbled the cables. The second attempt I used the transformer and got it up to 120VDC input. It failed while drawing arcs. I have a few more attempts left so I'm not worried about breaking a few IGBT's. Here are some pictures of my circuit and set up. I'm going to use the P-mosfets next time around to see how well it does. (forgive the armature-ish-ness of my windows paint edited circuit) 

EDIT: I added the TVS diodes though they are not marked on the circuit. They are in the picture just after the blue feed inductor going back to source. I'm not sure how to test them but I added 7 100v zener diodes in series to get a 700v rating. too high? should it be 600V?
Also bazar thing happened when I used the transformer. With only the transformer connected it began to oscillate very "weakly". It wasn't loud (I can still hear it, so under 20kHz) and the sparks off the secondary only formed on contact and did not stretch out more than, I don't know, 0.1mm. Super tiny but very unexpected. When I applied full power to the driver the pitch and volume of the oscillation changed and it back to big sparks. Not sure what that mean, but ill check in with the other modifications later.



Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 31, 2020, 01:28:43 AM
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 31, 2020, 01:29:59 AM
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on January 31, 2020, 05:50:52 AM

Have you switched to 1200V IGBTs now, or still using the 650V FETs?  I couldn't quite tell from your pictures.  The schematic still shows FETs.

BTW, have you measured your 25VDC gate supply voltage?  That would require roughly 18VAC from the transformer.

If you don't mind wasting some gate-drive power, you could skip the PFETs.  Old Dell laptop power supplies are often easy to find (I pull them from electronics recycling at work).  They are 19.5V.  Some 50-ohm 5watt resistors fed from 19.5V should work quite well, avoiding any need for PFETs.  (19V is across each resistor only ~50% of the time, so total power is roughly 0.5 * 19V * 19V / 50ohms = 3.61W.  Power will be a bit lower yet, since the forward drop of the gate diode plus the other IGBT will be more than 0.5V, so it's more like 17V across the resistor half the time.  1.5V for the other half of the time (into the gate zeners), which is almost no additional resistor power.

The peak IGBT collector voltage is nominally twice the center-tap voltage.  Thus 700V is too high for protection.  700V would be a good value if there were two such strings, one for each collector-to-emitter.  Or, ~400V for the center tap instead of 700V.  (If using 650V FETs, then even lower clamping voltage is needed.)

If those are standard zener diodes, they're not likely to survive high peak power as TVS diodes do.  (TVS diodes are optimized specifically for peak-power use.)  Not that it matters, but both are actually avalanche-breakdown devices, technically not actually zeners, even though we all call them zeners.

Concerning what voltage to use for clamping, don't forget that the UF4007 diodes are rated for 1000V.  Typical may be higher, but they could easily fail before 1200V IGBTs do.  I'd suggest using 300-350V worth of TVS diodes at the power inductor output, or 600-700V worth at the IGBTs.  That allows margin for their higher voltage during transient clamping events, while still handling the normal 530V IGBT peak (265V inductor output/transformer center tap) of 170V input.

Low-power oscillation occurs because the 25V gate supply feeds some current to the collectors through the gate resistors and diodes.  I've seen this before.  It can actually be helpful, having some oscillation running before the main supply is turned on helps reduce startup issues.  (I intentionally feed more bleed power to the main oscillator prior to full power.)

If you ever get a scope, I'd love to see the two gate-drive waveforms, and collector waveforms if you have probes to handle the higher voltage.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on January 31, 2020, 10:22:46 PM
Sorry, my mistake. I am using the 1200v IGBT's, part number IHW40N120R5XKSA1 and your right about the UF4007's. In both broken circuits I have at least 1 broken UF4007 and I think both zeners are dead (realized I was testing them wrong after the fact so I'll double check those zeners today).

The gate supply I used is just an old transformer connected to mains. I use a full bridge rectifier and a smoothing cap on the secondary and while drawing 100mA through the 220ohm resistor it reads 25VDC at the cap. With no load its a bit more voltage at the cap, but I cant recall how much. So I would still need a Zener diode unless I want to take some coils off the secondary, which I might. My other option is that I have a laptop charger from my old broken down computer. It puts out 16.5V and 3.65A Could this work with the appropriate resistor?

Good news on the P-fet front though! I did manage to get them to work. I had to keep the voltage at 18VDC to make sure I didn't damage the IGBT gates, but it worked. I'm going to order some 17V 5W zener diodes soon, but if they give me trouble I'm opting for the laptop charger or altered transformer. Good to have those options in place.

About the TVS diodes. They're cheap enough so I'm going to get some, but looking at the ratings has me all turned around again. Let me see if I understand this correctly. Normal operating voltage that the IGBT sees is pi x 170 = 534VDC. However from the power feed inductor to the source is half this value, putting it at 267VDC. That would mean I want a TVS diode with a working voltage as close to or slightly over 267VDC but a clamping voltage low enough to protect the UF4007, so 400 preferred to 500V max? Also, does it matter if they are uni or bi directional?

That's interesting about the low power oscillation. I think I' going to poke around at that a bit and will most likely include it as a start up switch in the final version.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on February 01, 2020, 04:51:52 AM
16.5V for gate drive can work.  There is some advantage to a higher voltage and zener clamping, however.  For a given current, the resistance is higher, so closer to a constant-current source.  That way the current doesn't drop as much at the IGBT's Vge plateau voltage, about 9V for your part at 40A.  Your 25V transformer solution could be fine, but 0.1A limits gate-drive current options.  Small 24DC supplies are somewhat common too - laptop style or even wall-warts.

Concerning TVS diodes, yes I think you understand.  The 1.5KE300 family should work well for the transformer center-tap, or two in series for the IGBTs collector-to-emitter.  Since they're fairly cheap, multiple sets in parallel is a good idea, to make sure the total energy stored in the power-feed inductor can be dissipated.  For this use, uni or bi directional are both fine.  Use whatever is cheaper.  (Uni-directional, of course, need to be wired the correct way around.  Bi-directional devices are just two uni-directional ones in series back-to-back in one package.  The extra forward-biased device slows down the speed a bit for very-fast voltage spikes.  ZVS circuits don't generate such fast voltage rise-times.)

Keep having fun!  And, good luck with your final voltage-multiplier use.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on February 07, 2020, 05:45:01 PM
Good news! It worked! I was about to give up and settle for less input voltage  when I decided to test something. If you remember I built my two E  sections of the transformer out of two UU sections glued together to form the E (4 U sections in total). On the central limb of the E I had the primary and on the outer limbs two secondaries in series. My gut told me there was something there so I separated the secondaries to test each one. I don’t know why, but one coil produced more power than the other. I just tested my drawing an arc at 60VDC input and one was hot and fiery while the other more blue and wispy. Testing the resistance of each coil showed they were about the same (16.6 vs 17.1 ohms) So my guess was that cheap amazon cores are to blame. I swap the coils around and sure enough its one limb gives more power than the other (forgive my lose use of the word “power”).

I decided to pull the whole thing apart and used just 2 U sections. I rewound a new primary and secondary for each limb and used acrylic to introduce a core gap of 2mm. From what I understand, core gap is one way to introduce leakage inductance and reduce coupling factor. No idea what mine is at, but it seems to work! Only problem I’m having is that at max input voltage the alligator jumpers and wires get so hot the plastic casing on them starts to melt! Easy fix though: more litz wire in the permanent set up.
 I no longer hear the oscillation and I can push it all the way to 170VDC input. For the gate drive I use 220 ohm resistors and found a different transformer that, after being rectified and filtered, put out 40VDC. I will continue to test and will be posting the complete constructing with pictures, parts, and links to videos once its done. I just need to look up how to build a soft starter because I’m worried If I were to just plug this thing into the wall it would just explode. It draws 14.5amps while pulling an arc at 120VAC input so it will trip the breaker if I used it for too long.

Thanks again for all your guys help.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on February 08, 2020, 05:13:57 AM
Great to hear of your success!

My soft-start is just an incandescent light bulb across the switch.  That gets oscillation started at low power as soon as line-voltage is applied.  A power resistor would work as well.  Another option that I might try next time is a diode from the gate-drive supply.  If your 40V transformer has enough current capability, you could do that, starting your ZVS oscillator at 40V before switching in 170V.

A single secondary winding is good too.  The two separate ones are less efficient even if working properly - more wire for a given voltage.  I'd guess that the issue was a defect in one of the windings - perhaps a shorted turn, rather than an issue with the cores.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on February 08, 2020, 04:42:46 PM
I forgot about that! Here I am thinking about getting relays and mosfets and all sorts of stuff to build this soft starter, but really all I need is a high wattage load to start up (like the light bulb) and then a switch to bypass it once its started up.

Your idea about the gate drive transformer for start up is interesting, but I'm not sure my transformer can supply enough current. Something I'll have to check out. If I did choose this route I would leave the negative wire connected to the emitter and then a diode from the positive to the "line side" of the power feed inductor (not sure what to call the side of the power feed inductor that attaches directly to the capacitor). Once started up I should just be able to turn on main power. Would the start up transformer be constantly supplying current long side main power? Should I toss in a start up switch to disengage the start up transformer once full voltage is running?
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on February 08, 2020, 06:29:10 PM
Yes, that's the correct connections for the gate-drive soft-start diode.  There's no need to switch it off.  That's what the diode is for.  Once line power is switched on, the ZVS input inductor is at 170V.  The diode blocks that 170V from feeding back to the 40V gate-drive supply.  (Without a diode, the gate-drive would fry at 170V.)

If your power is rectified line voltage without filter capacitors, then it will drop below 40V momentarily at each line voltage zero-crossing.  The gate-drive supply will feed a little bit of current at those zero-crossings (120 times per second).  Shouldn't be an issue for this case either.

If you go with gate-drive, I'd love to hear how it works.  I haven't tried that option yet - didn't think of it quite in time for my last ZVS project.  Expect it will work, and avoid a high-power resistor.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on February 10, 2020, 12:09:08 AM
More good news Dave! The low voltage gave drive soft start works! I had to make a few changes, probably will make more later, but as is:

The gate drive is rectified and smoothed with a capacitor. When just the gate drive transformer is pugged in 20VDC is seen at the capacitor, when drawing an arc it drops to about 18.5VDC. The soft start current idles at about 0.2A and tops out at around 1A while pulling an arc. When main power is applied, soft start current drops to 0A (as expected) and the gate drive voltage rises to 21.1VDC and only drops to 20VDC when drawing an arc.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on February 10, 2020, 05:29:33 AM

Wonderful!  I'm thrilled for you, and for the confirmation of what I plan to try next time I make a ZVS circuit.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 05, 2020, 06:10:36 AM
Its been a long while! I meant to upload all this sooner but life got busy. Now I have all the time in the world so I thought I should get back with the results :) I'm very happy with it and have more plans an modifications in store. Here are the only videos I have!

 I wanted to make better videos but my HV coil shorted through the oil and now it needs to be rewound. Ill get to it eventually but its not high up on the list. At the moment I am doing away with the center tap and using two inductors (opinions on this?). I want my driver to do double duty. One for HV stuff and the other as an induction furnace. I have the crucible and everything, but making it works as an induction furnace is not so easy. With the coil large I can use it to heat wrenches and screw drivers. I never tried to get them red hot because it was drawing 20+ amps and I'm working off a 15 amp breaker. With nothing in the coil it draws almost 15 amps using the same resonance capacitor bank as the HV project (I think 1.5uf?) but when I put something in the coil the IGBT's failed. So I thought I need to greatly reduce the capacitance. I drop it down to 10nF and it draws about 4 amps with nothing in it and 6 amps with a screw driver in it. But it failed again and now I'm waiting for more parts to arrive. I know its not HV stuff, but I thought I would share.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 06, 2020, 06:13:59 AM
Nice Jacob's ladder video - looks a lot like mine.  The first HV arc video looked impressive too.  Enough voltage to start an arc at that distance is going to stress your transformer.  Takes careful oil-filling under good vacuum to avoid air bubbles that generate corona and lead to failure.  I think kludesmith has good posts on vacuum filling.

Induction coils are generally to low inductance to run from a 120V-fed ZVS oscillator.  Most induction heater designs run from lower voltage higher current DC input, or use a transformer on the ZVS output to convert to lower voltage and higher current.

The other consideration for these ZVS oscillator circuits in general is that oscillation drops out of the resonant Q gets too low.  (Inserting the object to heat lowers Q.)  When oscillation drops out, current ramps up in the supply input inductor until something fails.  For the Jacob's ladder arcs, which also lower Q, the oscillation will switch to the higher frequency of the transformer's leakage inductance.  Such a scheme might work for induction heating too.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 06, 2020, 03:10:29 PM
Hey Dave! I defiantly drew some inspiration from you. Its cool stuff and mounting it on a piece of wood is very simple.

I don't have a proper pump for this stuff. I just use a fridge pump plus a little 5 volt pump to get enough out. I think the next one I make ill warm up the oil before hand to help with penetration. I'm also going to double up on the layers of paper in-between coils. Even so, I think the only reason it broke is because I was getting corona streamers coming off on of the transformer output terminals so I decided to see what would happen if I connected the other side to ground. The streamers got bigger, then it shorted and died... FOR SICENCE!

I had to look up what Q factor was and I think I understand. High Q factor means that there are many oscillations ringing naturally and thus the ZVS driver doesn't need to supply too much power to keep it going. However, when power is being drawn the Q factor is dampened and more energy is required to keep the oscillation going. A large enough draw could cause the oscillation to stop completely and you end up with a short in one of the IGBT's and ka-boom.  With the Jacobs ladder the reasons it continues to work is that it can fall back on the transformers no-load oscillation frequency. Do I have the right idea? It just seems odd to me that something should work at lower voltage and not higher. Low and high voltage seems like an arbitrary distinction when you talking a difference of 100V on the input. What I'm trying to figure out is why exactly it works on 60V every time no problems, but 120 its such a mess. Seems like I should just be able to get the properly rated parts and everything should work like normal.

The only reason I'm wanting the 120V ZVC induction heater is its simple and doesn't require a big transformer. I have a big transformer, but I like the challenge. In the end I have enough parts for 2 or 3 attempts before I call it quits and build 2 circuits. one for HV and one induction.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: John123 on May 06, 2020, 06:17:55 PM
Holy crap those arcs jump a pretty large gap!
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 07, 2020, 06:08:16 AM
Yes, ZVS oscillators stop oscillating when the Q is too low, which means the load impedance is too low compared to the resonant L and C impedances.  At low Q, the sine-wave doesn't ring down low enough to turn off the other IGBT, so it remains on and current ramps up.  Initially one IGBT is taking all the current as you said.  When the current ramps up enough, that IGBT's forward drop gets too high to keep the other IGBT off, so then both conduct.  Current continues to ramp until something fries.

Transformer leakage inductance limits how low the Q can get.  It's the transformer's shorted-output frequency, which is it's leakage-inductance, with the resonant capacitance, that keeps oscillation going.  That's a higher frequency.  No-load is the frequency that it starts at before an arc forms, which is a lower frequency (the transformer's full primary inductance, not just leakage inductance).

What sort of coil are you using for induction heating?  The issue is with normal copper-tubing coils used with internal water cooling.  Those typically have relatively few turns that are spaced from each other, so low inductance.  It's the low inductance that needs low voltage and high current (for a given frequency).  At high-enough frequency, the voltage goes up.  But the ZVS oscillator will have trouble as the frequency gets too high for the IGBT switching.  If your heating coil is more like an induction cooktop, then 120V will work fine as long as the Q is high enough.  Induction cooktops typically run with low Q, with an absorbing pan very close to the coil.  Adding a series inductance could fix that issue, behaving like a transformer leakage inductance.

Have you measured frequency with your induction heating experiments?  Of course, at lower voltage to measure without frying.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 07, 2020, 11:01:59 PM
Thanks John! They jump a maximum 14" and I estimate 400KV. I have plans for even bigger ones  ;D

Dave, this perfectly describes what I'm seeing when things fail. There is about a 75% chance only one IGBT will be shorted and broken. I normally save the others for tests later on. So I guess that's the one that gets stuck on and it dies before the other one does. When I use the salvaged ones they both fail 100% of the time things break down.

So if I'm understanding all this correctly, leakage inductance can be seen as an inductor in series with the primary coil. Its this leakage inductance that limits the current and how low the Q can get.  So all I should need to do I add an inductor in series with my heating coil? Seems simple enough.

I'm also realizing that the capacitor bank needs to be fairly large. When I used a much lower capacitance the frequency short right up. It was upward of 225kHz. For induction heating I guess you want to keep the frequency low, about 10-30kHz.

The coil I'm using is a 1/4" copper tube. Its not water cooled yet, because I'm still working out the kinks, but it will be eventually.

The one test I did with the low voltage was with IRFP260N MOSFET's and I was only watching the current. I made a small coil to put a screw driver in and as I did I watched the current creep up to 60 amps. 10 amps over its max rating at 25C for about 5 seconds blue them up pretty good. Even the heat sink was too hot to touch lol.

When the parts eventually come in (delays with shipping and all) I'll build and get back to you with more details. Well, as much as my DMM will tell me lol.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 08, 2020, 06:26:50 AM
Some induction heating works better at such high frequencies, especially heating more conductive metals such as aluminum.  The ZVS switching elements need to be able to handle the frequency.  Many IGBTs will struggle at 200+kHz.  So, for most uses, more capacitance for lower frequency is good.  But that makes the impedance lower, so more current at 120V direct input.

To prevent Q from getting too low, my simulations and experiments suggest 0.85 maximum coupling factor.  That's equivalent to a series inductance of 39% of your induction work coil if the series inductor is lossless.  For real conditions, I'd suggest a series inductor of at least 50% of the work coil inductance.

Still, if the resonant impedance is too low, the current can get too high just due to that low impedance, even when Q is still OK.  The IGBTs and power-feed inductors need to handle that current.  At 120V, you may end up with a very high-power heater and no way to turn it down.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 09, 2020, 09:42:12 PM
It works! I took a small ferrite transformer core and wound some wire around it runs smoothly :) Only issue was that the wires on it got hotter than the crucible lol. Some goes for the two power feed inductors. I assume that the losses you were talking about are in that heat? I switched out the 14 gauge wire with litz wire I made of 100 strands of 30 gauge wire. it runs considerably cooler now. I'm going to make the same modifications the power feed inductors.

I'm tweaking it to get the frequency higher as well. After some trial and error I have it drawing a little more than 10 amps and the tank circuit frequency is 41kHz. I know that with IGBT's the higher the switching speeds the more losses are incurred. I have them on a big heat sink with a cooling but when you say that at 200kHz they struggle, is that just with heat dissipation or it is a limiting factor for how quickly they can be switched?

Also, I got LTSpice. No idea how to use it, but I drew my circuit on it just to share. S1 is my low voltage turn on and S2 will not be powered unless S1 is on. Without the mandatory low voltage start the whole thing just dies.

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 10, 2020, 04:06:24 AM
Did you intend to have gate drive powered from the transformer output (left side of R1 and R2 to top of C1) rather than the high-voltage as shown?  Great that you have LTSpice.  With yahoo groups gone, I'm haven't looked where the LTSpice user group has migrated.  There's lots of information floating around on how to use it.  Start with it's built-in help.

Yes, you could probably get those IGBTs oscillating at 200kHz, but with increased power dissipation.

Great that you now have first-hand experience with the value of litz wire!  The power-feed inductors will be helped some too.  Since a significant fraction of their current is DC, litz won't make as much difference there.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 10, 2020, 06:11:05 AM
Woops! Your right! R1 and R2 should be connected before D5. The intention being that I'm driving the gates from the transformer output. That's how I have it on the workbench, just got to make that fix to the diagram. I also removed a series inductor.

I'll for sure be learning more about LTSpice, seems like it would be very helpful.

Its amazing the difference litz wire makes. The power feed inductors don't even get warm and the series inductor still gets hot, but I think it might be doable. I discovered that litz wire should be wrapped in tape though. A few turns on one of the power feed inductors got scratched and after running for 1.5minutes shorted through the inductor and blew an IGBT :'( The crucible got too hot to touch in that time so I'm hopeful that given more time it will get hot enough to melt copper and iron. Tomorrow will tell!
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 11, 2020, 12:09:53 AM
Yes, it's best to have an outer covering for litz wire.  Some comes with that already.

One more detail to fix on your schematic:  Now the bottom of L4 is on the low-voltage supply.  Once you learn how to simulate and plot node voltages, it makes finding such mistakes easy to do yourself.  I presume the actual construction is correct.

What core (if any) are you using for the series inductor L5?  Can you tell if the core is heating more than the wire?  Core losses can often be higher than copper losses.  BTW, have you measured inductance of L5 compared to L6 (your heating coil)?  If L5 needs to be 40% or more of L6 to prevent oscillation from dropping out at low L6 Q, but if it's high, say 100% or more, it may be hard to get enough power into L6.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on May 11, 2020, 04:49:58 AM
Woops! I didn't even notice it until you pointed it out but your right. L4 should be connected to the high voltage, not the low voltage. That's how it is on the bench, guess I should have taken a bit more time drawing that circuit haha.

The core does get hot! At first I used 9 turns on a PC40 transformer core. Litz wire was 100 strands of 30 gauge. The wire got warmish but after running for about 3 minutes the core was almost too hot to touch. So I thought maybe if I got another of the same core (I have three to spare) and super glue them together it wont get so hot. So I glued them together and doubled up on the litz wire (200 thread and managed i think 6 or 7 turns) and I was able to run it for 10 minutes before the series inductor got so hot I couldn't touch it. I've been trying to find out the solution to this problem and from what I'm thinking its just that PC40 isn't up to the task? I'm running at about 45kHz and 10 amps. maybe an air-core coil is the solution?

I don't have any tools to measure inductance so a lot of my work is trial and error. If I remove/add capacitors/coils from the series inductor how does it change the current and frequency? At one point I got it drawing 4 amps at a bit of 900kHz. I think I've struck a good balance, but is it possible that while I might be drawing 1200W a lot of it isn't getting to the crucible?

also, these are the cores I'm using for the inductors:

Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on May 12, 2020, 06:03:17 AM
When using ferrite cores for inductors (as opposed to transformers), they always need gaps to reduce effective permeability.  Otherwise the core is almost always saturated.  45kHz is a good frequency for PC40 material - looks like it's intended for roughly 25kHz-100kHz, although could be used a ways outside that range.

Don't worry about too-hot-to-touch.  PC40 is designed for 80-100C operation, with lowest loss at 90C.  If it's getting over 100C, then it's time to back off.  Most ferrites designed for power applications are designed to run hot (designed with minimum power loss at ~100C).

Do you have a scope?  Or, how are you measuring frequency?  If you have a scope, adjust your inductor gap until the inductor has half the voltage of the work coil with nothing inside the work coil.  Then your inductor is half of the work-coil's inductance.

You mentioned 900kHz!  That's impressive for IGBTs!  They are probably not too efficient at such high frequencies.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: Andrew321 on June 01, 2020, 03:38:05 AM
The core gap I really should have caught, that was half my battle getting the HV ZVS driver working in the first place lol. But that's really interesting about the ferriets working best at high temperatures. I had no idea they were optimized like that. I just thought that if its getting hot that means its wasting power and while a certain amount is acceptable/unavoidable, I thought the 100 degree range was just too much.
I don't have a scope, just a basic multimeter. I connect the leads across the capacitors just after the one closest to the coil. Seems to give me an accurate reading, but I don't remember the range its capable of so maybe the 900kHz was not quite spot on. I only had it running at the high a frequency for maybe 5 seconds or so. Just long enough to get a measurement.

I have given up on the HV ZVS induction heater for now. I almost had it prefect on my last set of parts. I didn't like the heat wasted in the inductor core so for the series inductor I tried the air core. It was 200 strand litz wire, 6 turns, 3 inches dimeter. I discovered I could plop a ferrite in and out of the coil (very small in comparison to the coil itself) to control the frequency and current some what. Nothing got hot. I was ready to actually try my first melt when I realized that the crucible wasn't sitting completely in the coil and the top 1/4 of it was sticking out the top of the work coil. I didn't think it mattered to lift the work coil so that it matched the crucible but when I made that one change, it fried the whole thing. I think in doing that I increased the coupling factor too much. I gave up on it for now and I set up a low voltage one using MOSFETS and my big step down transformer.
Eventually I'll come back to this induction heater. When I do I'll use properly rated MOSFETS (IGBT's get hot at 50-90khz after 15 minutes even with cooling fans) and instead of a separate inductor in series with the work coil I'm going to make the work coil longer than the crucible to reduce the coupling factor.

Thanks for all your help Dave! Don't think I could have done it with out you. When I get the high voltage stuff up and running again I'll make better videos and share.
Title: Re: Mazilli ZVS Driver Modification Problems
Post by: davekni on June 02, 2020, 04:53:24 AM
Adding extra turns is essentially the same as adding a series inductor.  It's simpler, but a bit more tricky to insure staying within limits for ZVS oscillation.  Sorry to hear of the frying event.  Encouraging that you have a likely cause.

If you can afford an inexpensive scope, that makes a dramatic difference in debugging and understanding circuit behavior.  Something like Hantek DSO5102P for $250 new on EBay from several stores.  I have the slightly more expensive version that includes a signal generator.  There are USB scopes that use a host computer for display and control, some a bit less expensive.  There are also old used scopes for yet less.

Keep learning and having fun!
SimplePortal 2.3.6 © 2008-2014, SimplePortal