Towards Distributed Control of Discrete-Event Systems

To initiate a discussion on the modeling requirements for distributed control of discrete-event systems, a partially-automated regionbased methodology is presented. The methodology is illustrated via a well-known example from distributed computing: the dining philosophers. 1 Decentralized, Asynchronous, and Distributed DES Control In this section, we explain our understanding of the gradual evolution of control for discrete-event systems (DES) from centralized control to more advanced forms, including distributed control, an area still in its infancy but one that may soon become vital for practical applications of the theory. Ramadge and Wonham’s theory of non-blocking supervisory control for DES was introduced in [23] and developed soon after in [18, 24] to cover situations in which only some of the events generated by the DES are observed, for instance “in situations involving decentralized control” [24]. The basic supervisory control problem consists of deciding, for a DES with observable/unobservable and controllable/uncontrollable events, whether a specified behavior may be enforced on the DES by some admissible control law. This basic supervisory control problem is decidable. Moreover, if every controllable event is also observable, then one can effectively compute the largest under-approximation of the specified behavior that can be enforced by the supervisory control. After the work done in [28] and extended in [35], decentralized control refers to a form of control in which several local supervisors, with different subsets of observable events and controllable events, cooperate in rounds. In each round, the local supervisors compute a set of authorized events and the DES performs authorized events until one is observed by some local supervisor, which puts an end to the Work partially supported by the EC-FP7 under project DISC (Grant Agreement INFSO-ICT-224498) and NSERC.