A General Architecture for Decentralized Supervisory Control of Discrete-Event Systems

Abstract

We consider a generalized form of the conventional decentralized control architecture for discreteevent systems where the control actions of a set of supervisors can be``fused'' using both union and intersection of enabled events. Namely, the supervisors agree a priori on choosing``fusion by union'' for certain controllable events and``fusion by intersection'' for certain other controllable events. We show that under this architecture, a larger class of languages can be achieved than before since a relaxed version of the notion of co-observability appears in the necessary and suf®cient conditions for the existence of supervisors. The computational complexity of verifying these new conditions is studied. A method of partitioning the controllable events between``fusion by union'' and``fusion by intersection'' is presented. The algebraic properties of co-observability in the context of this architecture are presented. We show that appropriate combinations of fusion rules with corresponding decoupled local decision rules guarantee the safety of the closed-loop behavior with respect to a given speci®cation that is not co-observable. We characterize an``optimal'' combination of fusion rules among those combinations guaranteeing the safety of the closed-loop behavior. In addition, a simple supervisor synthesis technique generating the in®mal pre®x-closed controllable and co-observable superlanguage is presented.

Knowledge Based Scheduling of Distributed Systems

“…A distributed controller sets up such a supervisor per each process. In a conjunctive supervisor [15], in order to execute an enabled transition t that belongs to several processes, all the corresponding supervisors must agree to fire it. In a disjunctive supervisor, it is sufficient that at least one of the supervisors allows (supports) t Instead of constructing supervisors, one per processfor single or sets of processes, we transform the processes, represented as sets of transitions in a Petri Net.…”

“…In this way, we will be able to related the sets of states of the original and transformed version. The supporting process policy can be classified as having a disjunctive architecture for decentralized control [15]. Although the details of the transformation are not given here, they should be clear from the theoretical explanation.…”

Distributed Control Synthesis

“…This (finite state) automaton observes the controlled system, progresses according to the transitions it observes, and blocks some of the enabled transitions, depending on its current state. In a similar way, in distributed control [34,26,25], for each process we assign such a supervisor, which changes its states each time the process it supervises makes a transition, or when a visible transition of another process (e.g., through the change of shared variables) is executed. Based on its states, the supervisor allows (supports) transitions of the controlled process.…”

“…Based on its states, the supervisor allows (supports) transitions of the controlled process. In a disjunctive control architecture [34], if no supervisor supports an, otherwise enabled, transition, it cannot execute and is thus blocked. Such a supervisor can be amalgamated, through a transformation, into the code of the controlled process.…”

“…A related problem, which we address in this paper, is to automatically design control to distributed systems [25,26,34]. Synthesis is achieved in an incremental way: an already existing distributed system is modified to satisfy an additional property.…”