Model-based fault diagnosis using nonlinear estimators: a neural approach

Published 1 January 1997

A. Alessandri, Thomas Parisini

DOI10.1109/acc.1997.609657

Citations20

TL;DR

The use of neural networks is introduced as reliable functional approximators, thus allowing an on-line application of the proposed FDI scheme, exploiting a novel class of nonlinear discrete-time sliding-window observers.

Abstract

The problem of model-based fault detection and isolation (FDI) is addressed. An architecture for FDI is devised, exploiting a novel class of nonlinear discrete-time sliding-window observers. Theoretical motivations for such an architecture on the basis of convergence properties of the observers are addressed, and a rather large class of faults are considered, including actuators, sensors, and several kinds of plant malfunctions. Moreover, the use of neural networks is introduced as reliable functional approximators, thus allowing an on-line application of the proposed FDI scheme.

Keywords

Engineering

AutomaticaFault diagnosis in dynamic systems using analytical and knowledge-based redundancy

3,481 Citations1990Paul M. Frank

The principles and most important techniques of model-based residual generation using parameter identification and state estimation methods are outlined with emphasis upon the latest attempts to achieve robustness with respect to modelling errors.

IEEE Transactions on Automatic ControlObserver design for nonlinear systems with discrete-time measurements

336 Citations1995P.E. Moraal, Jessy W. Grizzle

It is shown that the discrete Newton method, properly interpreted, yields an asymptotic observer for a large class of discrete-time systems, while the continuous Newton method may be employed to obtain a global observer.

IEEE Transactions on Systems Man and CyberneticsAutomated fault detection and accommodation: a learning systems approach

305 Citations1995Marios M. Polycarpou, A. Helmicki

In this framework, neural network models constitute an important class of on-line approximators and adaptation/learning schemes and a systematic procedure for constructing nonlinear estimation algorithms is developed, and a stable learning scheme is derived using Lyapunov theory.

Control Engineering PracticeKeynote paper: Fault diagnosis and neural networks: A power plant application

54 Citations1995Guido Guglielmi, Thomas Parisini +1 more

Simulation results obtained by a very accurate and validated model of the plant show the effectiveness of using multilayer feedforward and Radial Basis Functions neural networks to solve real fault-detection and diagnosis problems.

IFAC Proceedings VolumesA Survey of Observer Based Residual Generation for FDI

21 Citations1994Vasanth Krishnaswami, Giorgio Rizzoni

A survey of dynamic system health monitoring schemes that are based on the theory of linear and nonlinear observer design, including the Unknown Input Observer, Eigenstructure Assignment and Detection Filter schemes is presented.

On convergence of neural approximate nonlinear state estimators

6 Citations1997Thomas Parisini, A. Alessandri +2 more

A min-max technique is proposed to determine the weight coefficients of the "neural" observer so as to estimate the system state to any given degree of accuracy, thus guaranteeing the boundedness of the estimation error.

Neural approximators for nonlinear sliding-window state observers

3 Citations2002A. Alessandri, Manfredi Maggiore +2 more

State estimation on the basis of noisy measures is discussed, and a convergence condition for the state estimate is given, and neural approximation of the nonlinear observer is considered.