An SCR?! You're... not going to like that control response, I think... Also it's drawn as a schottky? And the gate is on the wrong side? Huh, weird library.
I'm assuming CT is from half (or full) bridge inverter output (to main power transformer)?
What's FED? Seems to be something negative polarity, but symbol just shows resistor...
Gate drive transformer doesn't look right either -- when an output pulls low, the NMOS is still fully on (by about 12V worth). It's biased for short circuiting If you're doing a complementary source follower, it needs to be biased class AB, not, erm, class A-self-destruct.
The emitter followers also I think aren't doing anything, or much; the 3525's outputs are strong enough for this I think. They're certainly not going to make it switch any faster (the outputs are slew rate limited to 200-300ns, in my experience; obviously, followers can't improve that any). It's even capable of driving transformers by itself; but in this case you probably do need some boost, for the couple ~kW worth? of inverter transistors.
So, best way to start a design like this, is current mode.
Current mode, current mode, current mode.
Current control protects the inverter transistors, gives you current limit 100% for free, avoids filtering / pole compensation irritations, allows the loop to run much faster -- huge value.
If this is stick/TIG, you don't need voltage control anyway, besides maximum output (which can just be maximum PWM, it doesn't need to be a regulated voltage). If you want to replicate "anti stick" (whatever that is -- I'll have to look it up, what kind of characteristic that needs), or target MIG as well, you'll probably need a lower impedance, so either mixed control (perhaps constant-resistance (Thevenin equivalent)), or CC/CV operating curve (i.e. put down a voltage error amp, to drive the current mode control's input).
You can still have a peak current detect and latch for protection purposes; could also implement desat protection, to a similar end. Not a bad idea anyway, as current mode control is a bit rough from the primary side -- notice the inverter current is pulsed, it's only carrying current while on -- so the sensed (average) current is proportional to PWM%, reading significant error at low PWM%. Probably a mix of peak detect and averaging is good for the current transformer sense network.
You don't want entirely peak current, by the way -- for feedback. It's impossible to compensate, it'll slide into a chaotic mess. Nonlinearities in the control loop are a huge problem. Admitting even just a softer peak-detect function may prove tricky. The problem is the asymmetry of rise and fall times, and the little squigglies on your waveforms (from the inverter current itself, and in what the CT senses -- inevitably it's going to have some ringing, etc.).
The best solution is secondary side current sensing -- at these currents, this is easiest done with a Hall effect sensor, and, since you're using a gate drive transformer (GDT), the controller can be secondary side referenced, making voltage sense trivial, and making a current sense shunt resistor an option as well. (Just beware of the inductance of a resistor that size, and the wire up to it. Some RC filtering, and differential sensing, will be required.) Sensing current directly at the output filter inductor (preferably the output side just before the filter caps, or the rectifier ground return) gives you the continuous inductor current waveform, no interruptions -- unlike the CT which only senses while the inverter is switched on. Thus, no peak-detect hacks needed.
Tim