Failure Detection and Isolation in Integrator Networks

TL;DR

This paper extends failure detection and isolation techniques from Laplacian consensus networks to general linear networked dynamics, including bidirectional and undirected links, with validation on large networks.

Contribution

It introduces new FDI methods for general linear networks, handling bidirectional and undirected topologies, with scalability and directionality insights.

Findings

FDI techniques effectively detect link failures in large networks.

Directionality impacts failure detection performance.

Scalability of methods demonstrated on large network experiments.

Abstract

Detection and isolation of link failures under the Laplacian consensus dynamics have been the focus of our previous study. Our results relate the failure of links in the network to jump discontinuities in the derivatives of the output responses of the nodes and exploit that relation to propose failure detection and isolation (FDI) techniques, accordingly. In this work, we extend the results to general linear networked dynamics. In particular, we show that with additional niceties of the integrator networks and the enhanced proofs, we are able to incorporate both unidirectional and bidirectional link failures. At the next step, we extend the available FDI techniques to accommodate the cases of bidirectional link failures and undirected topologies. Computer experiments with large networks and both directed and undirected topologies provide interesting insights as to the role of directionality, as well as the scalability of the proposed FDI techniques with the network size.