Grinding ferrite is surprisingly easy. While the surface is quite hard, the inside is noticeably softer.  Found this the hard way while grinding down the centre leg of an E core.

Grinding is is used in the final finishing of ferrite core sets - how else would they get such tight tolerances on mating faces.

Diamond tools are quite unnecessary - corundum and carbide are both hard enough to cut ferrite.

I strongly recommend taking the time to build jigs to hold your ferrite, and position it accurately for whatever shaping process you use. Get the jig perfect, then making matching pieces becomes easy.

An LC filter might be more effective.
Have you measured the current spike? Knowledge is power...

I agree with TMaxElectronics: just add the resulting actuator signals.  A properly tuned PID should be robust enough to deal with any funnies.  I would be inclined to make one of the PID loops slower than the other though - I have no math to back this up, just a gut feeling that its going to be more stable that way.

Ideally you'd want to put a circulator/isolator between your power amp and your work head, so that the inevitable reflected power can be dumped into a dummy load.  Not sure where you'd get one that can handle 1kW, and it probably wont be cheap.

Some of those scope traces look like you are trying to drive the comparator input below its negative rail (evidence: soft clipping).  That would explain the weird output from the AD8055. Post a schematic of what you are doing.
Also I wouldn't bother with a simulation unless you are using a proper spice model of the comparator/opamp - try either LTSpice or Simetrix, they will give you much more accurate results. Both are free, Simetrix is node limited but is my personal favorite due to ease of use and simulation speed.

[quote author=johnf link=topic=1768.msg13419#msg13419 date=1632893967
I meant to say that tuned circuit "Q" is made better by increasing capacitance with reducing inductance ie "Q" is proportional to "C" in a tuned circuit
[/quote]

That may be somewhat true in practice, buts its not because Q is proportional to C (its not). Its because reducing L to keep the same F_res also tends to reduce the R in the tuned circuit (unless you have really shitty capacitors).  Reducing R is the real reason the Q rises.

I'm glad you found FEMM to be useful. Numerical simulation (done with the appropriate caveats in mind) is great for getting answers where analytical solutions are prohibitively difficult. Analysing magnetic systems is definitely one of those situations.  I've been fortunate enough to have access in the past to Ansys Maxwell, a full 3D electromagnetic FEA software package. Very capable, very slow on complex problems (days to run some of my simulations), and of course very expensive (NZ$30k per CPU core!).  I was using it to simulate PCB coils for wireless charging, so the geometry was very complex, hence the long simulation times.

Eagle? Eww! Just spent 6 months forced to use that POS. And the paid version too. Never again.

Are these drivers a single IGBT quasi-resonant topology like induction stoves?

It would seem that way based on the schematics I've seen. The only power devices in this inverter are the one IGBT and the diode bridge.  With a single switch pushing 1kW I can't see how else it could be done - the way the transformer is built would lead to considerable leakage inductance, and the only sensible way to handle that, at that power level, would be a quasi-resonant topology.

So far everything is still running, after cooking dinner last night. And I still have a set of spares if it dies again.

And like most appliances these days I guess its designed to outlast the warranty, and not much more.  So they can sell you another one.
https://en.wikipedia.org/wiki/Phoebus_cartel <- this is why consumers get screwed over, and the world has a waste problem.

Odd.  Is Q2N3904 really a PNP?  2N3904 is very common NPN.  Can't find any data on Q2N3904.

try 2n2904