Design and analysis of supervisory controllers for discrete event dynamic systems

A supervisor is an agent that is capable of asserting certain constraints on the uncontrolled plant behavior, in response to a series of events generated by the plant. Given the uncontrolled plant behavior and the control specifications, the problem is to design a supervisory controller such that the plant and the controller together will satisfy the specification. In practice, the plant and the supervisor are distinct entities. Neither the plant nor the supervisor alone describe the controlled behavior of the plant.; In this thesis, we synthesize the supervisor and the plant models to obtain a combined model. This combined model describes the controlled behavior of the plant. It can also be used for the analysis of the controlled system, the verification of the supervisor and the control code generation. Since our methodology separates the plant design specifications from the controller specifications, it is not needed to reconstruct the system model from scratch in case the control requirements are changed.; Construction of the combined model has three major steps: Design of the plant model and the supervisory controller, separately, and the synthesis of the two models. The method is hybrid because PN and R&W framework are used for the plant and supervisor design, respectively. Since the supervisor is a finite automaton, we propose some reduction rules and an algorithm to reduce the size of the supervisor. This way, the complexity of the combined model is also decreased.; The reduction algorithm and the rules, as well as the synthesis algorithm for the combined model are independent of the R&W framework and are applicable to any supervisor represented as a finite automaton. To this end, we present an alternative supervisory controller design for a class of problems. Unfortunately, the reduction procedures may not always reduce the supervisor to a manageable size. Hence, we introduce a decomposition and reconstruction methodology to obtain a PN controller from a supervisor with finite automata structure.; The existing supervisory controller design techniques revolve around the functional behavior of the system. For that reason, time related specifications cannot be covered effectively. We explore this issue in the second part of the thesis. We consider a plant model where state dependent conflicts and free choices are possible. The plant is modeled as a timed Petri net. We propose a solution methodology to design a dispatcher that sequences the activities while optimizing some performance measures. Even though the problem is a scheduling and sequencing problem, there is a peculiar similarity between this problem and the supervisory control problem: The dispatcher acts like a supervisory controller and the performance measure is analogous to a control specification.