The testing would consist of running the diesel generators under load but not operating in abnormal conditions.
I’ve built and programmed control systems that do all this with multiple generators, multiple utility feeds, and multiple tiers of loads of different priorities.
You can test the system’s ability to detect loss of power or poor quality power by lifting sensing wires or injecting out of frequency or voltage range power with a test set, and load pickup and load testing we used a load bank. The actual hospital load is fed through bypass breakers during these tests so if the utility went out we would have to manually operate breakers to disconnect the load bank and swing the vital loads over to the generators.
If I am setting up electrical protection on a generator I don’t put a short circuit on the generator, I show the protection relay what it would see by injecting current and voltage with a relay test set and make sure the relay gives the signal to open the breaker when it should.
Greenfield hospital generation sites are straightforward to open the utility feeds since the vital loads are still fed through the old power system.
In the end you WILL test what happens when the hospital plunges into darkness. Would you prefer the first test to be a a bright sunny day when all the staff is ready for something (and you can turn the utility power back on quickly if something fails), or when someone with a backhoe/chainsaw has an accident with your power feed. The latter will happen, I don't know how or when, but at some point in your future trees will take down power lines, backhoes will cut something, major weather events will take out power for days, maybe a large blackout...
You should have course test everything separately often. However if you wouldn't let a random electrician (electrician only because they will know how to not kill themselves) turn off your utility connection with only a day of warning, your system isn't trusted to handle real events which are typically much worse than a clean turn off the switch.
FYI where I’ve worked the tiers of loads are metered individually and picked up in order of priority and with a check to verify the combined capacity of the generators which are online is great enough to pick up the load. Likewise loads are shed in order if there isn’t enough generation such as one generator tripping.
I don’t really see that much difference between opening a breaker and bad quality power. It is the same sensing and logic in both cases: is the magnitude and frequency of the phase to phase voltages in the acceptable range or not?