Premium
Online Bayesian model assessment using nonlinear filters
Author(s) -
Kontoroupi Thaleia,
Smyth Andrew W.
Publication year - 2017
Publication title -
structural control and health monitoring
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.587
H-Index - 62
eISSN - 1545-2263
pISSN - 1545-2255
DOI - 10.1002/stc.1880
Subject(s) - kalman filter , computer science , bayesian probability , nonlinear system , online model , inference , bayesian inference , recursive bayesian estimation , noise (video) , algorithm , hierarchy , approximate inference , data mining , machine learning , artificial intelligence , mathematics , statistics , physics , quantum mechanics , economics , market economy , image (mathematics)
Summary Model assessment is an integral part of many engineering applications, because any analytical or numerical mathematical model used for predictive purposes is only an approximation of the real system. The Bayesian approach to model assessment requires the calculation of the evidence of each candidate model considered given the available measured data, which is a nontrivial task, and it is usually attempted offline, e.g., by using a stochastic simulation scheme or some deterministic approximation. Very few authors, in general, and hardly any in the field of structural dynamics, have investigated online application of model assessment. The current work explores how Bayesian model assessment and an online identification scheme for joint state and parameter estimation, in particular the unscented Kalman filter, whose computational efficiency has been widely recognized, could be integrated into a single method. This hierarchical Bayesian modeling approach involves two inference levels, namely, model assessment and parameter estimation. There is the possibility of adding another level within the hierarchy for noise estimation. An illustrative example involving several hysteretic candidate models is presented to demonstrate the implementation of the proposed procedure in structural health monitoring applications. Copyright © 2016 John Wiley & Sons, Ltd.