As a stress engineer who has performed lots of FEA, most of which has been in support of certification and some of which has been in support of failed tested items, I can confirm that ElectroVlasic's option 4 is basically how it works.
If you take the example of the wing, if you can prove the load path and boundary conditions of the area surrounding the failed component (for example by test data from strain gauges and the like) then you can generate a finite element model of another component of whatever design, and providing the interfaces are the same apply the known loads and boundary conditions and bob's your uncle. The certifying authorities need to be convinced that your models are correct, and if they are then no further physical testing is required.