Hi,
That's hard as the 110 to 120 V AC is a fluctuating source voltage. I'd maybe sample the rectified 110 VAC (and thus avoid the need for a dual supply or an additional circuit block to rectify or block the negative part of the 110V AC for the watchdog circuit - although even a simple reverse-biased diode at the lower resistor of the voltage divider for the V AC would keep input signals to a diode drop below ground) and feed it into a comparator via a voltage divider with a reference voltage on the non-inverting input with appropriate hysteresis for expected rectified ripple margin.
If the input to the comparator falls below Vref, its output should go low and in this way it can de-assert whatever controls that input source's pass device (MOSFET, relay, whatever...) and assert the 28V source's path. It's very simple logic: if A is high, B must be low, if A goes low B must go high. A simple NOT gate at the comparator output to complement and invert the comparator's output signal can be used for the Q OR !Q method of: If A = 1 then B = 0, if A = 0 then B = 1.
If you expect over-voltage peaks that might damage the control circuitry (the comparator and any logic devices), then you'd need to clamp the sampled input voltage with, for example, a precision clamp to somewhere between Vref and the comparator's supply voltage.
A large hold-up capacitor is needed at the junction of source A and source B where they meet the load, based on time it would take from A going low to B going high.