Health monitoring for strongly non-linear systems using the Ensemble Kalman Filter

Abstract

Many structural engineering problems of practical interest involve pronounced non-linear dynamics the governing laws of which are not always clearly understood. Standard identification and damage detection techniques have difficulties in these situations which feature significant modelling errors and strongly non-Gaussian signals. This paper presents a combination of the ensemble Kalman filter and non-parametric modelling techniques to tackle structural health monitoring for non-linear systems in a manner that can readily accommodate the presence of non-Gaussian noise. Both location and time of occurrence of damage are accurately detected in spite of measurement and modelling noise. A comparison between ensemble and extended Kalman filters is also presented, highlighting the benefits of the present approach. Copyright © 2005 John Wiley & Sons, Ltd.