A failure detection and isolation algorithm for a decentralised multisensor system

Abstract

This paper describes an algorithm for detecting and isolating failures and its application on a multisensor decentralized system. We consider decentralized systems as a network of local processors where communication is performed only between neighbouring nodes. Local observations are taken by each node, and using the information form of the Kalman filter, local information contributions are calculated. These are then communicated and a global estimate is computed. The failure and isolation technique described in this paper is based on the use of multiple models, which describe the changes on the system dynamics when a failure is produced.<>