Big change since last update. The most important part? LLC sucks. Back on page 4 I said it sucks using no capacitor at all. Adding a matching inductor keeps the tank cap from blowing transistors, but it does it at the expense of reactive current. Power factor is still low, typically 0.3. That means for the 10kW output I want, I need 30kVA of inverter output. That's better than 100kVA, but it's still a lot of trouble to go to, and costs a lot of efficiency.
So, contemplating things one day, I decided to try series resonance instead. Whereas the tank cap shorts out the inverter when switching a parallel tank, in series the coil handles all the harmonics, so switching is easy. At resonance, power factor is exactly one, and it's very easy to find because it's just the classic RLC circuit analyzed by every EE student in the world. I looked for the largest ferrite core on hand, a rather oversized flyback transformer core measuring about 3μH/T2 and saturating at 10AT.
This was the setup as I had it, basically the same circuit on the previous page, nailed down to a plywood board. You can see the transformer, bottom left of center, with about 40 turns 8AWG primary and 6 turns 1/4" copper tubing secondary. This did not work well, because leakage inductance was bigger than the work coil's inductance! I really should've put the primary on top of the secondary, but I didn't have much wire, either. A better winding is definitely needed. Toroidial transformers are excellent for this...
I calculated I need about this much transformer, having the fortune to find a couple of these rather large high-mu ferrite toroids for very cheap at All Electronics. (As of 12/07/09, it seems they don't offer them anymore.) I measured one at 6μH/T2 and 10AT saturation, needing four to run down to 10kHz at 160V supply with 10 turns primary, one secondary. I forced it into the existing setup, shorting the work coil and tank capacitor together, with this thing inbetween, with the primary connected directly to the inverter output (still with coupling capacitor).
The result? Immediate success. Whereas before, I was strugging to get 700W power output from 240V supply with large reactive inverter currents (> 30ARMS), right away I got over 1000W from 120V supply with unity power factor!
Here are some waveforms. Above resonance, current is low and risetime is long (I still had an output snubber connected at this time). Closer to (but not at) resonance, current is larger, and since Q is high, phase is still nearly 90° (note: current transformer was backwards, so the waveform is upside down). When Q is low enough to safely reach resonance (without drawing dangerously high currents), current slides right into phase and power output goes way up.
Next step, simplify. I doubled the number of turns on the output transformer, and doubled the minimum frequency spec to 20kHz, cutting the required inductance down by a quarter, allowing me to use just one core. I soldered together a new tank cap of 100 x 0.1μF MKPs (these gray caps are cheaper and smaller than the blues from before, and don't handle nearly as much current!).
Moving from proof of concept to prototyping, I put this together. From right to left: the breadboard handles front panel functions (buttons, controls and lights); main control board contains power supply, DC-DC converter, startup timer, overload latch and oscillator (feedback circuit not yet implemented); the two gate drives on individual boards; the nearer heatsink hides the output transistors, with coupling capacitors, supply capacitors and bridge rectifier behind; and the output transformer, with tank capacitor and work coil on the far left. Take a video tour:
And as I note in the video, I really ought to have a control circuit on top of that thing. That was the very next thing I put together. The circuit is very similar to earlier circuits, with these changes: feedback can be entirely inverter-side, so instead of taking tank voltage, inverter current is monitored with a current transformer. Amplitude feedback is now in terms of current (also from the current transformer), and instead of voltage, current phase is measured relative to the inverter's output. (For isolation's sake, I get the inverter phase (minus a few nanoseconds propagation delay) from the oscillator's output.) A type II phase detector is used to ensure sufficient range (the XOR detector used earlier probably won't actually work, and definitely won't work with the input phases chosen here.) I got rid of the gate drives and DC-DC converter, since I'm building a fairly small unit here (I'll save those for the 10kW model). Instead, the oscillator (a TL494) drives a pulse transformer which drives the IGBTs directly. To compensate for the desat protection the gate drives had, a peak current cutoff was added. The general block diagram is shown below.
With the circuit together and working, I put it in a fancy new chassis -- heavy aluminum, to reflect the magnetic fields inside, whereas steel would absorb it and heat up.
Here's the inside view. Output network on the left, with cooling fan. The inverter board mounts on the center divider (which keeps strong magnetic fields on the left side and switching noise on the right), while the control board mounts on the right hand wall. Rectifier and filter caps are also mounted to the divider. A line filter sits on the bottom.
And here's the look with the lid on. Yeah, it's obviously handmade, but it's not a disaster either, it worked out fairly well considering it's what I had on hand. Notice the flare connections on the left side to connect to the work coil, and the holes for water hoses in front.
And here's what it can do. Now, you can't really tell because the camera doesn't read the brightness very well, but this is a graphite disk about an inch across being heated to yellow hot (you can tell the color by the glow on and around the coil). It took about a minute to get there, drawing pretty close to 1000W, the inverter running at about 60V and 20A RMS, PF = 1.0 (the half-bridge inverter supplies a 160Vp-p square wave, which has an RMS fundamental component of about 60V which actually drives the tank). A stack of pennies placed between two wads of ceramic wool will take a few minutes to melt (it goes faster if you use a graphite crucible, or preheat them because hot copper is fairly resistive).
Update 5/09/2010: Here's video of a properly built model. This is more like a production model! It was built with homemade PCBs, (almost) all brand-new parts, and the chassis made according to plan using 0.06" aluminum stock.