Probably not cost effective as opposed to either making the software more resilient or just letting some monkeys/interns loose in the cockpit and pressing everything (With some focused searches obviously)
I don't know, I suspect software resiliency is one of those "gets exponentially more expensive as you approach perfect" situations and a robot could bash inputs at speeds a human couldn't in thousands of orders, it would certainly let you test those weird code paths that blow up when they are never exercised until they are.
It does get exponentially more expensive, which is why you have to pick a certain level of ~perfection~ and then stop.
Also, the robots would have to bash at the speed of a human: The point is to mimick human inputs not necessarily test every code path (Which you could do in software)
When testing resiliency software testing isn't the whole of the thing.
I spent a lot of time at new job (well 18mths now) on making the software side more resilient, we still had a site wide outage of internet access and comms and internal systems because one of the women in the sales office unplugged a socket in the ingress comms cabinet to plug her hair straighteners in.
Coincidentally my request to have physical access control on all the comms cabinets shot right up the list.
The boundary layer between software and hardware itself has to be tested, the real world is a messy place :).
