I would say that even for arbitrary wave-form the firmware for this should be entirely feasible if you simply do:
simulate this circuit – Schematic created using CircuitLab
The optocoupler can be an opto-triac like the ones in the MOC series, but more aimed at the 48V-ish regions, or just an Opto-MOST rated correctly. For the latter you might have to increase R4, but that's all down to datasheets.
You can even put the entire system between the rectifiers to trigger a SCR (in stead of a triac) with just a simple MOST, but that will cost you 1.8V to 4V of the top of your 50VAC wave, since the load will also trigger through the rectifier (for mains that's not mush of an issue, but here it's an actual percentage).
Of course, you would be best off (least noise and interference) to handle each phase separately, but with the right uC math, if you do this for one phase, you can also control the other phases, if they come from the same generator with known phase angles (presumably 120 degrees).
Also, don't forget snubberage for inductive loads. Snubberage is a cool word, for that alone you should just always do it!
The interrupt in a 8MHz RC-running 8-bit MCU can be handled quick enough, if done right, so even with software lagging in reactive triggering, you'll turn on in microseconds, but if the frequency change is not in 100's of Hz per seconds, you can also design a very good predictive triggering scheme that can get you right on the 0 every time, apart from the first 3 crossings in a frequency change or so. But that's harder on the FW side.
Obviously: keep the resistors such that the current dumped into VCC through R2/D1 is always smaller than what R1 drains when the MOST is on (since the MOST turns on well before its gate reaches VCC+Vf, R1 will always be draining current when power is being dumped into VCC). Keep the resistors as small as is safely allowable, for faster switchy-times. Of course, you can increase R1 to 10k if you know the MCU will ALWAYS drain more than the dump into VCC from your 50VAC through R2/D1, which makes the signal even sharper (though, how noticeable?)
Anyway, enough for you to puzzle on, me thinks.
EDIT: Of course, your MCU has some sort of input somewhere from the main control, likely optically isolated.
EDIT: If you do the SCR inside the rectifier trick, you can do it all with two or three transistors again and no MCU, but as said before: voltage loss.