HYBRID MODELING TECHNIQUES AND THEIR APPLICATION TO FAULT-TOLERANT COMPUTER SYSTEMS.

We detail the use of behavioral decomposition in modeling systems that contain state transitions having widely disparate time constants. We show that such decomposition leads naturally to hybrid system models, containing both analytic and simulative submodels. We propose a language of Extended Stochastic Petri Nets (ESPN) for use in specification of the simulative submodels. We illustrate salient features of our approach, especially the flexibility of the ESPN specification, with applications to the modeling of ultra-reliable flight control systems, that is, systems for which reliability requirements are likely to exceed 1 - 10** minus **9.