State observation and diagnosis of discrete-event systems described by stochastic automata

The problems ofstate observation and diagnosis are solved for discrete-even t systems, which are described by stochastic automata. As many systems are not observable in the sense that it is possible to reconstruct the state un ambiguously, the observation problem is set up as the problem of determinin g the smallest possible set of states that are compatible with the measured input and output sequences. The diagnostic problem is shown to be, in prin ciple, an observation problem. Conditions for the observability and diagnos ability of stochastic automata are presented. The results are illustrated b y examples.