I inherited a large and fairly complex state machine. It has 31 possible states, all are really needed (big business process). It has the following inputs:
I can't think of any easy way to do test an FSM like this with out getting really pedantic and employing proofs, using machine learning techniques, or brute force.
Brute force: Write a something that will generate all the 4320 test cases in some declarative manner with mostly incorrect data. I would recommend putting this in a CSV file and then use something like NUnits parameteric testing to load all the test cases. Now most of these test cases will fail so you will have to update the declarative file manually to be correct and take just a sample of the test cases randomly to fix.
Machine Learning technique: You could employ some Vector machines or MDA algorithms/heuristics to try to learn on the sample you took from what we mentioned above and teach your ML program your FSM. Then run the algorithm on all the 4320 inputs and see where the two disagree.