This is the hard part of the UML specification. In the simplest case, when you enter a state containing orthogonal regions, the initial psuedo state in each orthogonal region essentially triggers a separate control flow. There are many complex rules about how events are consumed by these threads and how threads join together.
But, in the opinion of the methodologist I highly recommend (HS Lahman), you really shouldn't use more than the usual old Moore startup machines , for more information on why you should use Moore state machines (which you can model well in UML), and not in Miley or Harel state machines, see this passage from Lachmann's Book . For more information on the difference between a Moore machine and a Mile machine, see this question at StackExchange .
source share