A normally open switch won't keep the feedback capacitor charged when the output is off. The gate needs to be connected to source with a double throw switch. You are probably getting inrush sometimes.
I have parts on the way for my next version, this time for sure! I'm staying with sequential contacts. I'm convinced that's the best solution for a general purpose gadget to connect and disconnect 116vdc with load current happening. It avoids the need for a switch and allows me to use it when I don't have access to the pre-contactor voltage.
I'm giving up on hard switching, too many unknowns and blown fets to continue that approach. I don't know the real impedance of my inverter. And I can't even measure it, but I think the actual inrush is what's killing my fets, hundreds of amps, could even be 1,000. Datasheets seem to say that's ok for a few microseconds but the results say otherwise. I still believe that a 0.1uF cap is a low impedance gate drive. It has less voltage rise than a 100ohm gate resistor for any amount of current for up to 10uS and will absorb transients better as well. But it's academic, I'm going back to linear feedback.
So here's my new (sort of) circuit. I found a way to reduce the parts count of the sequential ramp circuit I posted previously using only one low power fet. And I've found ways to suppress the nasty little transients that happen on connection. This will ramp up in about 1 second with less than one amp even if the inverter is on and drawing 50 watts idle.
I've carefully selected a new fet but will need two to get the low on resistance I want. The pair are slightly more capable than the big single as far as soa, power and current, and not as expensive. They are 150v but that should be ok, I hope. https://www.mouser.com/ds/2/205/DS100228A(IXFH160N15T2)-347934.pdf There are very few mosfets that can meet the specs I need at this 28s voltage. Or at least that's what the datasheets say, or don't say.
I'll build and test this in a few days, maybe this will be the design that really works.