8kW CW DRSSTC driven by ZVS (Mazzilli) oscillator

[davekni]

After a couple successful ZVS oscillator designs at ~1.5kW from rectified 120VAC line, decided to attempt a more powerful ZVS using rectified 240VAC line.

Chose a CW DRSSTC as initial use for this ZVS oscillator.  Not a great choice in hind sight.  More on that in a second post.  Reused secondary from my initial 2014 CW SSTC (which was previously bottom-fed with no primary).  Added a center-tapped 5-turn copper pipe primary.

   

/>
Schematic for ZVS oscillator:



Basic oscillator topology is same as standard Mazzilli ZVS design.
    IGBTs U1 and U2 are the power switches.
    D1 and D6 are the cross-coupled feedback diodes.
    L1 and L2 with 0.8 coupling are the 5-turn DRSSTC primary coil.
    C1 is primary resonant capacitor (MMC).
    L3 is DRSSTC secondary.
    C9 simulates top load capacitance.
    C13 and R7 simulate arc load.
    L5 is the power inductor feeding center tap of primary coil.
    D6 and R6 power low level oscillation when line voltage is zero.  This has been discussed in other ZVS oscillator threads as an aid to startup.

Required several tweaks to basic Mazzilli ZVS oscillator circuit to function on rectified 240VAC line at 8kW:
    1)  1200V 75A IGBTs instead of FETs (STGYA75H120DF2).  Measured Vce breakdown is 1360-1380V.
    2)  TVS diodes to clamp voltage when load drops.  D7+D9 clamps voltage directly across power inductor L5.  D9 absorbs most of the energy within L5 when current drops, more efficiently than direct IGBT Vce TVS diodes.  To handle wiring inductance, Vce clamps are still necessary.  D3 and D14 clamp U1 and U2 respectively.  (IGBTs generally handle less avalanche energy than do FETs.  TVS clamping is therefore more important for ZVS oscillators using IGBTs.)
    3)  6VDC gate level shift (offset) to accommodate increased Vf of 1600V feedback diodes and IGBT Vce at high current.  This consists of zener D2, C3, R9 for gate of U1 and zener D4, C2, R13 for gate of U2.
    4)  Emitter follower gate drive buffers.  Q3, Q4 for U1 gate and Q1, Q2 for U2 gate.
    5)  Even with above emitter follower buffer, reasonably high pull-up current is helpful for feedback diodes.  Functioned initially with just 500 ohm pull-up resistors R4 and R5.  Adding coupled inductors L7 and L8 in series with pull-up resistors increases pull-up current at critical switching times without increasing power dissipation significantly.  Clamp diodes D22 and D23 are needed to avoid voltages above 24V due to inductors.
    6)  Series resistors between feedback diodes and emitter followers are needed.  Due to parasitic inductance and IGBT diode forward recovery time, IGBT collectors can spike negative to ~-30V.  Series resistors prevent spikes from damaging emitter followers.
       6a)  Probably not necessary, but above series resistors are split into turn-on and turn-off values.  Unlike bridges, turn-on needs to be fast.  Turn-off resistance is 406 ohms (R2 and R3).  Turn on adds parallel 130 ohm resistors R1 and R11 through diodes D18 and D19.
    7)  Also probably not necessary, but I added a bit of capacitance across Vce, C6 and C8.  Idea is to reduce Vce spikes due to interconnect inductance and switching times.
      Final patch was the hardest to figure out and perhaps the hardest to explain.  Unlike FETs, IGBTs have a turn-on delay even if Vge is applied well before Ice current is applied.  ZVS oscillators switch nominally at zero volts.  Collector current is ideally a step function at switching.  Supply current through L5 ideally switches instantly from one IGBT to the other.  Even with Vge on early, Vce spikes as Ice rapidly rises.  Without any added circuitry, that Vce spike briefly turns on the IGBT that had just turned off.  Causes excess IGBT heating and can even cause a couple cycles of oscillation at switching.  My previous 1.5kW 120VAC units managed OK without too much trouble.  Became more of a problem at 240VAC with peak currents above 100A.  Added circuitry clamps Vge in the off state for a while after each on-to-off transition.  Circuitry is in dashed boxes on schematic, near bottom on left and right sides.

Here's a simple LTSpice simulation showing this IGBT turn-on delay Vce spike.  Uses crude IGBT simulation consisting of an NFET and PNP BJT.  Vge is constant +15V.  Ice transitions from 0 to 100A across simulation times 10ns to 20ns.  This crude IGBT model happens to be faster than these actual parts, resulting in a shorter spike than actual.  But the general shape matches exept for time scale.  Red trace is Vge, constant 15V.  Green trace is Vce showing 40V spike.  Orange trace is Ice.




For completeness, here is a schematic of the control board.  It turns the ZVS oscillator on and off at line voltage zero crossings.  Also prevents ZVS operation until 24V gate supply is up (UVL).  Control circuitry here is implemented with discrete transistors as I often use, so not anything others would want to copy:



My conclusion is that ZVS (Mazzilli) oscillators can be made to work from rectified 240VAC line.  Should be fine for induction heating (except that it provides no isolation of line power).  For the linear resistive load of induction heating, this design would likely function well to 15kW, and perhaps to 20kW.  All my 8kW DRSSTC testing and video was done without fans connected.  IGBT copper heat spreader over heatsink increased from 22C at start of testing to 31C at end of testing.

[davekni]

Now back to my unfortunate choice to use a DRSSTC as a test use for this higher power ZVS oscillator:

Biggest advantage of ZVS (Mazzilli) oscillators is their simplicity.  Obviously, all the above tweaks required for 240V high power operation defeat that simplicity advantage.

ZVS oscillators have a second advantage,  but only for some uses.    Typical H-bridges provide a constant voltage in series with a series-resonant L/C circuit.  ZVS oscillators provide a constant voltage across a parallel-resonant L/C circuit.  Constant parallel voltage is ideal for induction heating.  Power drops when load is removed.  A series driven induction drive requires cycle dropping or some other feedback control to maintain constant coil voltage and current (to prevent current from becoming excessive when load is removed).  Even with cycle-dropping, IGBT current remains high since resonant current remains high.  On the other hand, ZVS IGBTs conduct very little current when lightly loaded even though resonant current remains high.

This ZVS oscillator should work well for driving an HV "flyback" transformer too.  That's my plan next for this ZVS oscillator, a higher power weed-zapper.

However, constant parallel voltage is NOT a good choice for DRSSTC primary drive.  Arc loads tend to be somewhat constant-voltage.  As input power increases, arc extends to consume additional power without significant increase in top load voltage.  My ZVS oscillator tracks rectified line voltage.  Once input voltage is high enough to start arc, increased volage forces rapid arc growth and power increase.  Makes for bad line power factor in spite of having no bulk capacitor.

Even worse is if arc grows enough to drop secondary frequency below primary frequency.  Oscillation transitions from lower pole frequency to upper pole frequency.  Between those too frequencies is another resonant "zero" frequency, perhaps could be called an anti-resonance.  Primary Q drops way low at this anti-resonant frequency.  Not an issue if series driving primary with an H-bridge.  Power drops momentarily.  With parallel drive, low Q becomes high load.  ZVS input current spikes quite high.

Below is an LTSpice AC sweep showing the two poles and zero between them.  Zero is what causes high current with parallel drive of primary.  Solid green trace is magnitude of R2 current (1VAC supply current).  Dashed green line is phase of same current.  Phase crosses 0 at three resonant frequencies.  First is at ~81kHz of lower pole.  Third is at ~121kHz of upper pole.  These two frequencies have high Q so low current for this parallel drive topology.  The middle frequency of 100kHz is the "zero" or anti-resonant frequency.  Q is low, resulting in high current for parallel drive.



Here's a scope capture of ZVS input current (green trace at 5A/div) and Vce of one IGBT (red trace at 50V/div).  Peak current is ~135A.  Second plot is zoomed in to see frequency transition from lower pole at ~75kHz to upper pole at ~115kHz.





Above traces come from initial testing of single line half-cycles.  Power is through isolation transformers.  Reduced line voltage to 215VAC using transformer taps.  Transformer impedance limits current some.

Added a ballast resistor in series with 240V line before continuous operation.  This prevents damage due to high current and sudden current drop after pole transition.  Ballast resistor is R14, 0.65 ohms at 2kW.  Also increased primary capacitance (MMC) a bit to stay away from pole frequency transition.  Original intent was 10kW from 240V 40A circuit.  Reduced to 8kW due to ballast resistor.  RMS line current is still high, ~56A, due to poor power factor.

Below are scope captures of rectified line voltage after ballast and rectified line current.  Captures show roughly one line cycle (two half-cycles).  This is running in my garage with ground foil 1.25m above top of breakout, same setup used during video.  Line voltage (after ballast) is red trace at 10V/div.  Line current is green tract at 10A/div.  Both are rectified so positive only.  First capture is before arc strikes ground foil:



Second capture is later in burst with arc striking grounded foil:

[Anders Mikkelsen]

Impressive job getting such a temperamental circuit to perform at these voltages and power levels!

How sensitive is this circuit to parameter variations? Do a lot of values have to be modified if you use different IGBTs, change the operating frequency, coupling or primary impedance?

Did you have a chance to scope the IGBT collector current? With a low impedance tank like this across the two collectors, any conduction overlap will lead to the transistors conducting the instantaneous tank current. The best way I found to deal with this is to use a series blocking diodes, a controlled degree of conduction overlap, and some phase advance on the switching to make it switch before voltage zero crossing, but my experience is with induction heaters with a Q in excess of 100, and at higher frequencies. This is analogous to using diodes in parallel with the switches in a voltage fed inverter together with dead time and phase lead. I investigated current fed (technically inductance-fed as the source impedance is high at Fsw and low for DC) for the advantages you mention.

Interesting observation on the effect of the constant voltage behavior of the inductance fed push pull leaading to bad mains power factor. Could potentially be addressed by forcing the double resonant system to operate at the zero frequency between the poles, if this would be practical or possible. I know people tried operating voltage fed DRSSTC tanks at the zero frequency, and found it to exhibit the constant voltage behavior that the inductance-fed drive exhibits at the poles.

[klugesmith]

Excellent pushing of the envelope, Dave!
With exploration of the kinks, and presentation thereof.

Can we see some photos of the hardware? (Circuit boards etc.)
Tell us more about your 0.65 ohm 2 kW ballast R.

I have never made a "high power" inverter.   First application in my list is to run a tanked X-ray cathode or anode power unit from a Bennett system, originally working at 100 kHz with external resonant capacitors.  You got me thinking about trying a Mazzilli driver.  Would start with 10 or 20 volts instead of 240 volts.

[davekni]

Impressive job getting such a temperamental circuit to perform at these voltages and power levels!

Thank you!  It became an obsession.

How sensitive is this circuit to parameter variations? Do a lot of values have to be modified if you use different IGBTs, change the operating frequency, coupling or primary impedance?

I'd originally used AOK30B135W1 IGBTs, 1350V rated (~1700V measured), intended for resonant applications.  Don't have any turn-on times in specification, only turn-off.  Was very slow turn-on with large Vce spike even with Vge on early.  Lots of testing with short (<1ms) enable times.  Fried on second half-line-cycle of real testing.  This was before figuring out a solution for Vce spikes turning on opposite IGBT.  Peak currents are also higher than I'd initially expected (bad power factor as mentioned above).  AOK30B135W1 have much lower peak current capability.

Changing IGBTs might also need changes to feedback resistors to optimize turn-on and turn-off times.  Wasn't that critical with this use, but would be much more so with high-Q load.

Most marginal (I think) remaining design issue is voltage window between IGBT Vce breakdown (1360V) and TVS clamp voltage.  In above pole frequency transition, Vce peaks at 1340V with sudden power drop.  IGBTs with actual Vce breakdown closer to 1200V might or might not survive such an event depending on their internal avalanche energy capability.

Frequency is fine at both poles and probably any lower frequency.  If frequency extends too high, would need to reduce R/C time constant of Vce spike protection (circuit that prevents opposite IGBT from turning on).

For DRSSTC drive, it is sensitive to coupling and primary impedance.  Lower impedance or more coupling would rapidly increase power.  Opposite for lower coupling or higher impedance.  For other uses, shouldn't be so sensitive.

Did you have a chance to scope the IGBT collector current?

No, that is difficult to do without adding excessive parasitic inductance.  Do you have suggested techniques?  Not sure I could even add a tiny Rogowski coil without adding problematic inductance.

The best way I found to deal with this is to use a series blocking diodes,

I did consider such.  Perhaps due to one of your previous posts.  Might be more necessary for high-Q induction load.  Of course, diodes need to be faster than IGBT switching to be useful.  Without diodes, IGBT would desaturate a bit turning turn-off.  Not sure how that switching power would compare to diode losses.

Interesting observation on the effect of the constant voltage behavior of the inductance fed push pull leaading to bad mains power factor. Could potentially be addressed by forcing the double resonant system to operate at the zero frequency between the poles, if this would be practical or possible. I know people tried operating voltage fed DRSSTC tanks at the zero frequency, and found it to exhibit the constant voltage behavior that the inductance-fed drive exhibits at the poles.

Interesting.  Makes sense.  I hadn't heard of anyone running at the zero frequency before.  Both UD2.7... style feedback of H-Bridges and this ZVS circuit naturally lock at pole frequencies.  I think it would require some active control to force operation at zero frequency.

[davekni]

Excellent pushing of the envelope, Dave!
With exploration of the kinks, and presentation thereof.

Thank you!

Tell us more about your 0.65 ohm 2 kW ballast R.

I'd found and purchased three 1.3 ohm 2kW resistors on ebay.  They were quite delayed in coming.  So I kludged a 0.65 ohm ballast resistor for testing:



Made from scrap around my garage.  Thin SS sheets folded to be stiff enough form end terminals.  Small shear cuts along sides and top hold nichrome ribbon wedged into cuts.  NiCr ribbon glows orange and sags every time coil runs.  BTW, two of the strands broke after being hit by a secondary strike.  This was near the end of my testing and video, with spinner on top of top load.

When the three 1.3 ohm 2kW resistors finally came, they were a bit larger and much heavier than I'd expected.

Can we see some photos of the hardware? (Circuit boards etc.)

Sure.  Coil assembled, but without top load.  The ferrite plates on top are visible, part of my first 2014 SSTC design.  Secondary has 0.4mm PP sheet wrapped around it for storage as varnish coating is very thin on this coil.  It is 600mm high, 160mm diameter, 27AWG, 78mH.  Primary center tap is on left of image.  Not physical center of this 5-turn primary because upper turns are larger diameter so have more inductance per turn.

 [ You are not allowed to view attachments ]

Closer up view from right side of previous image.  Shows connection to primary.  Leads from MMC are 0.1mm copper foil.  Paper clamps connect foil to secondary pipe.  I'd ran early testing with no secondary and longer runs to verify clamps didn't overheat.

 [ You are not allowed to view attachments ]

With coils removed, view into driver and MMC.  Built into a standard 4-gallon plastic milk crate.  MMC is made from 2.7nF 1600Vdc PP caps on ~1mm polycarbonate sheet with copper foil tape.  Main ZVS power input inductor was an old surplus inductor that I modified.  Cut winding into 8 sections and wired in parallel.  Actually two sets of 4 connected separately to the two paralleled bridge rectifiers in order to force better current sharing between bridge rectifiers.

 [ You are not allowed to view attachments ]

View directly down from top:

 [ You are not allowed to view attachments ]

View of back side of MMC.  Shows how additional MMC sections are added for tuning.  Added sections are also on 1mm PC and copper foil tape.  One side of top extends with 127um polyimide insulation and foil tape.  Folds over top of main MMC and clamps on with paper clamps.

 [ You are not allowed to view attachments ]

I have never made a "high power" inverter.   First application in my list is to run a tanked X-ray cathode or anode power unit from a Bennett system, originally working at 100 kHz with external resonant capacitors.  You got me thinking about trying a Mazzilli driver.  Would start with 10 or 20 volts instead of 240 volts.

Would likely work well.  I've made quite a few lower power FET ZVS oscillators.  Several drive HV transformers.  Just make sure transformer coupling is less than 0.86 so that it can handle shorted load (or startup charging of HV load etc.).  ZVS frequency transitions from transformer leakage inductance at high load (shorted secondary) to transformer parallel inductance at lighter load.
Also make sure primary resonant capacitors are on same board with FETs.  Needs low parasitic inductance from FETs to primary cap(s) since current ideally switches instantly from one FET to the other.

[davekni]

Made this ZVS (Mazzilli) oscillator base into an HV weed zapper.  Made a HV transformer to replace TC top.  No ballast resistor needed for this load, so power reaches just over 10kW when load impedance is optimum.  Output voltage is +-12kV when input line voltage crests at 330V.  Youtube video of zapping weeds and moss in my back yard:
   

/>
This HV transformer runs at 25.5kHz with 529nF capacitance and no secondary load, resonating with transformer's primary parallel inductance.  Increases to ~36kHz with low impedance load, resonating with transformer's leakage inductance. 

How sensitive is this circuit to parameter variations? Do a lot of values have to be modified if you use different IGBTs, change the operating frequency, coupling or primary impedance?

Low frequency required two tweaks, neither of which would prevent higher frequency operation as well.  Discovered first issue in simulation:  Filter capacitor after bridge rectifier (C5 + C14) needed to increase from 940nF to 11uF.  Otherwise resonated with power line inductance.  Second change became obvious in use.  Core of input inductor L5 measured 180C after finishing above video.  Probably was a bit hotter yet before I got to measuring.  Rest of circuitry was plenty cool.  I've now added a fan to L5.  It had no active cooling previously.  For extended use changing to a ferrite-core inductor would make sense.

[alan sailer]

Your weed burning reminds me of an old comic line,

"I'm not a vegetarian because I love animals. I'm a vegetarian because I hate plants."

If you decide to turn this into a gopher killing device I will worship you.

Cheers.

[davekni]

If you decide to turn this into a gopher killing device I will worship you.

That reminds me of another story from someone I knew at work long ago.  He decided to deal with gophers by filling hole with acetylene and igniting it.  Turns out that the gopher tunnel extended under his house.  Explosion cracking foundation.

The high-voltage relatively-high frequency could be used for ozone generation.  Ozone sent down a gopher hole (and presumably spread through tunnels) could be enough to be lethal, which I gather is ~50PPM for small animal 2-hour exposure.  I think typical air-fed ozone generators make ~1% ozone.  Probably easier to feed CO2 down hole.  A container of dry ice with tube from container to gopher hole.  Of course, I have no idea if either option would be at all effective.

BTW, weed electrocution is used in some organic farming and at least experimentally for killing weeds along sidewalks.  One company in the Netherlands was advertising a high-frequency device for safety (and lighter weight).  Most of the farming units are 50Hz or 60Hz powered by generators pulled behind a tractor.  Here's a couple video links:
   

/>   

[Twospoons]

As I understand it electrocuting weeds works better than flaming, as the electrical heating extends right down the plant stem into the roots, instead of just burning off the top and leaving the roots intact. I've been meaning to turn an old MOT into an electro-weeding tool.

[davekni]

As I understand it electrocuting weeds works better than flaming, as the electrical heating extends right down the plant stem into the roots, instead of just burning off the top and leaving the roots intact.

That's my understanding too.  Key is killing roots.  To work well, top of soil needs to be reasonably dry to force current farther down through roots.

I've been meaning to turn an old MOT into an electro-weeding tool.

Use extreme caution!  MOTs regularly kill, including a man here in my small town ~20 years ago.  I have several MOTs, but am too cautious to consider them for weed zapping.

BTW, from what I read, commercial weed zappers are 8kV to 15kV.  I'd guess the 2.5-3kV of an MOT would be sufficient with a bit more dwell time on each weed to heat root sufficiently.

[Twospoons]

Don't worry, I'm quite aware of how dangerous MOTs are. Same as any high powered device with more than a few hundred volts on it. Electric weed killers are not something you just hack together -  one of the reasons I haven't done it yet. I'm still thinking through the kinds of safety interlocks I want, how everything can be insulated, how I'd make a safe handle etc.  As for the  'low' 2kV voltage, the usual diode doubler will help there.