Hmm, it's also a C divider, and loaded with a nice low resistance.
I wonder if it's actually protection against capacitors drying out. Really unusual.
See, if it were dV/dt, it should be one series cap, and a modest size parallel resistor. In quasi-steady-state ramping, Vbe = 0.6V and I = Vbe/R = C * dV/dt. If the 100nF were absent, this would give 1.35V/us, an unimaginably fast rate for a DC power supply with, I assume, electrolytic capacitors on the output.
But with the C divider, we get a pole-zero transfer function, where at low frequencies, the response is asymptotically a differentiator, but at high frequencies it shelves to a constant 2/102 ratio. For high frequencies to activate the base, it would need Vbe * 102/2 = 30V. This would be on the order of the 100n * 221R time constant, or shorter than 22us (frequencies above 7.2kHz).
This would basically only be plausible if the output filter capacitors went completely open circuit.
I'd think a more sensitive circuit would be desirable if the concern is detection of capacitor aging. For example if the output ripple is nominally, say, less than a volt, but we want to detect and disable output if it's more than, say, several volts: then the series resistor and capacitor would still be used, but:
- Instead of a zener as shown, a 1N4148 from GND to base would absorb reverse charge
- Instead of 100nF B-E, a smaller value (or the 2nF being larger, and probably the 1.5k being smaller), say 4.7 to 10nF
- Instead of 221R B-E, a larger value, say 10k, just to provide some ground reference without much loading the signal (this will give some dV/dt control as well)
- An LED could be connected in series with the collector, giving visual indication of the protection circuit activating.
But this even assumes anyone will look at the power supply ever, and not simply replace it whole, as is typical for maintenance (and that's in turn assuming the equipment itself is worth servicing at all). I suppose it might've been appropriate back in the day, but it's very hard to justify something like that today.
Today, a "health monitor" sort of function might be useful; you could sample the error amp level, output ripple, input ripple (through an opto -- optos have poor gain consistency, so this would be a very rough figure, but that's okay), output current and voltage, and maybe some temperatures (ambient / on-board, heatsink, etc.), and just put limits on them and forget about them for the most part. Maybe have them viewable with a service or debug mode. But when a limit is exceeded, log an error, or report to the user or cloud or whatever, so that appropriate action can be taken.
Which, as you can see, is quite a bit more complicated, but it's potentially the same sort of idea, just on a different scale. Most computers, test equipment, automobiles, etc. do this, at least in a broad way (POST and such). Expensive things -- expensive enough that it's worth spending the extra time to implement these features, to save on service time later.
Tim