Showing papers by "W. M. Wonham published in 2010"

Supervisor Localization: A Top-Down Approach to Distributed Control of Discrete-Event Systems

Abstract: We study the design of distributed control for discrete-event systems (DES) in the framework of supervisory control theory. We view a DES as comprised of a group of agents, acting independently except for specifications on global (group) behavior. The central problem investigated is how to synthesize local controllers for individual agents such that the resultant controlled behavior is identical with that achieved by global supervision. In the case of small-scale DES, a supervisor localization algorithm is developed that solves the problem in a top-down fashion: first, compute a global supervisor, then decompose it to local controllers while preserving the global controlled behavior. In the case of large-scale DES where owing to state explosion a global supervisor might not be feasibly computable, a decomposition-aggregation solution procedure is developed that combines the supervisor localization algorithm with an efficient modular control theory.

On the Computation of Natural Observers in Discrete-Event Systems

Abstract: Natural projections with the observer property have proved effective in reducing the computational complexity of nonblocking supervisory control design, and the state sizes of the resulting controllers. In this paper we present an algorithm to verify this property, or if necessary to achieve it. A natural projection is a special type of general causal reporter map; for the latter an algorithm is already known for verification and modification. This algorithm could be used to verify the observer property of a natural projection, but if the natural projection is not an observer the algorithm is not applicable to modify it to an observer. Also, while a general reporter map always admits a unique smallest refinement with the observer property, a natural projection does not. Indeed there may exist several minimal extensions to the original observable event set of a natural projection. We show that the problem of finding a minimal extension is NP-hard, but propose a polynomial-time algorithm that always finds an acceptable extension. While not guaranteed to be minimal, it is in practice often reasonably small.

Supervisor localization for large-scale discrete-event systems

Abstract: We propose a top-down approach, called supervisor localization, to distributed control of large-scale discrete-event systems (DES) The essence of this approach is to allocate, or localize, ‘external’ supervisory control action to individual plant components as their ‘internal’ control strategies To cope with state explosion of large DES, we develop a decomposition-aggregation procedure that employs a flexible decentralized and hierarchical architecture which reduces computational effort in localization This procedure is demonstrated in detail on a benchmark Production Cell that has state size of order 108