Communication Optimizations for a Wireless Distributed Prognostic Framework

Distributed architecture for prognostics is an essential step in prognostic research in order to enable feasible real-time system health management. Communication overhead is an important design problem for such systems. In this paper we focus on communication issues faced in the distributed implementation of an important class of algorithms for prognostics – particle filters. In spite of being computation and memory intensive, particle filters lend well to distributed implementation except for one significant step – resampling. We propose new resampling scheme called parameterized resampling that attempts to reduce communication between collaborating nodes in a distributed wireless sensor network. Analysis and comparison with relevant resampling schemes are also presented. A battery health management system is used as a target application.