A parallel sensor scheduling technique for fault detection in distributed parameter systems

Abstract

The problem under consideration is to determine a scheduling policy for discrete scanning sensors in such a way as to maximize the power of a simple parametric hypothesis test, which verifies the nominal state of the considered distributed system specified over a given multi-dimensional spatial domain. The optimal activation schedule for sensor network is determined based on the D s -optimality criterion defined on the respective Fisher Information Matrix. A computational scheme based on the branch-and-bound method is provided for the solution of a resulting combinatorial problem. Then, the parallel version of solver is developed in order to overcome the excessively growing computational costs for large-scale problems. The proposed approach is illustrated by simulations on a sensor network scheduling problem for a two-dimensional convective diffusion process. © 2008 Springer-Verlag Berlin Heidelberg.