On the supervisory control of discrete event systems

This thesis discusses a model for the supervisory control of discrete event systems. The purpose of discrete event system models is to describe those aspects of a plant which cannot be adequately modeled within the paradigm of difference or differential equations. The present investigation is based on a state transition model for the open-loop behavior with separate means of control; this is the model originally proposed by Ramadge and Wonham (RW-Model). For our discussion we adopt an extension of the RW-model to infinite sequences of events rather than finite strings.; The RW-model distinguishes between two classes of events: events that can be disabled and events that are always enabled. A generalization of this notion of control is proposed for the analysis of large-scale discrete event systems with interacting components. We derive necessary and sufficient conditions for the solution of an important supervisory control problem and present a polynomial-time algorithm for synthesis of the corresponding supervisor.; We use the extended model for an analysis of Boolean coordination problems for multi-component systems. Necessary and sufficient conditions for the existence of a solution in the absence of synchronization are given. It is shown that the general coordination problem can become computationally intractable. We extend the analysis of coordination problems to the case of discrete event processes with synchronization which we model with shared events. Sufficient conditions for the existence of a solution to basic synchronous coordination problems are derived. We demonstrate the significance of the result by showing that coordination problems can become computationally intractable if these conditions are violated.; We present an extension of the RW-Model to infinite real-time behaviors, which is based on a model for finite real-time behaviors suggested by Brave and Heymann. It is shown how to solve a basic supervisory control problem under the assumption that the supervisor can react arbitrarily fast to events occurring in the plant. Finally, we study the effect of delay in the communication between plant and supervisor. Necessary and sufficient conditions for the existence of a supervisor which implements a desired closed-loop behavior are given. The complexity of the corresponding synthesis algorithm is investigated.