Supervisory control of discrete event dynamical systems with partial observations

In this work, we first define sufficient observation and work spaces for supervisors of a class of Discrete Event Dynamical Systems (DEDS) in the Ramadge-Wonham framework. A sufficient observation space is a collection of events whose observation by a supervisor is enough to realize a given desired behavior. A sufficient work space is a collection of events (sufficient observation space) for which the control action of the supervisor is limited to the controllable elements of the work space. We construct algorithms to evaluate a sufficient work (or observation) space with a rather small cardinality. Reduction of work (or observation) spaces results in reducing the event detectors, communication and command channels between the supervisor and the plant. The algorithms are based on testing a normality and an observability condition. We also develop two effective polynomial time algorithms to test the normality and the observability conditions for regular languages.;Next, we study the Supervisory Control and Observation Problem (SCOP) of DEDSs where some events are not detected. Given a desired closed loop language K, the SCOP has a solution if and only if K is both observable and controllable. It is well known that the supremal observable and controllable sublanguage of K, i.e. the optimal solution for SCOP, does not exist if K is not observable. By introduction of an extended normality property, we construct a suboptimal solution for the SCOP. We prove that our suboptimal solution contains, in the set theory sense, those reported in the literature and thus is an improved suboptimal solution for the SCOP.;Then, we introduce the concept of partially decentralized supervisory control of DEDSs when the control objectives are locally specified. We also study the problem of decentralized supervisory control of DEDSs when the control objective is given in terms of predicates defined on the state space. We introduce notions of statically and rationally decentralizable predicates. When a predicate is decentralizable, we construct decentralized supervisors that can realize the predicates that define the control objective.