A Particle-Filtering Based Approach for Distributed Fault Diagnosis of Large-Scale Interconnected Nonlinear Systems

TL;DR

This paper introduces a distributed fault detection method for large nonlinear systems using particle filtering and consensus algorithms, enabling real-time fault diagnosis with communication among nodes.

Contribution

It presents a novel distributed particle filtering approach combined with consensus algorithms for fault diagnosis in large-scale nonlinear systems.

Findings

Effective fault detection demonstrated through numerical simulations.

Distributed approach reduces the need for multiple estimators.

Communication topology based on graph theory enhances information sharing.

Abstract

This paper deals with the problem of designing a distributed fault detection and isolation algorithm for nonlinear large-scale systems that are subjected to multiple fault modes. To solve this problem, a network of communicating detection nodes is deployed to monitor the monolithic process. Each node consists of an estimator with partial observation of the system's state. The local estimator executes a distributed variation of the particle filtering algorithm using the partial sensor measurements and the fault progression model of the process. During the implementation of the algorithm, each node communicates with its neighbors by sharing pre-processed information. The communication topology is defined using graph theoretic tools. The information fusion between the neighboring nodes is performed by means of a distributed average consensus algorithm to ensure the agreement over the value of the local likelihood functions. The proposed method enables online hypothesis testing without the need of a bank of estimators. Numerical simulations demonstrate the efficiency of the proposed approach.