Okay, built and tested. I found a sacrificial mosfet in my junkbox for the test.You could look at the average (filtered) voltage across the MOSFET.
If it stays at zero for a significant length of time, indicating a shorted MOSFET, then you could use that to open the relay.
Below is the simulation of a circuit to do that.
The NO relay contact is closed when there is power and a PWM signal at the MOSFET drain.
The PWM source signal is filtered by R1C1 and goes low with a delay [V(1)] after the simulated PWM gate signal stops and goes high (simulating a shorted MOSFET).
This turns off transistors Q1 and Q2, opening the relay and power to the motor some 9 seconds later.
As long as the PWM signal average is at least 1.5V, the relay will stay pulled in.
C1 is connected to the +12V so that [V(1)] will go high and energize the relay when power is first applied.
(Note C1 polarity).
D1 clamps the V(1) voltage to ≈13V maximum (about 0.7v greater than the 12V supply).
View attachment 120869
The relay engages as soon as power is applied, and switches off when
power is removed. For the test, I increased the pwm until I was at max rpm,
then I increased the load until the mosfet popped, & then I started counting
the seconds. After I got to 15, I yanked the plug because I didn't want to wait
any longer for the protection circuit to kick in, and having either my motor or pump damaged.
It was really screaming.