Vector Autoregressive-based Structural Identification Method by Means of Bayesian Inference

Abstract

Modal identification involves the determination of natural frequencies, damping ratios, and mode shapes of a mechanical system using measured vibration data. The vector autoregressive (VAR) method and its variants are popular techniques capable of quickly extracting the modal properties, whose parameters are entries in the system matrices and are estimated by linear regression. However, those methods originally provide only the best estimates of modal parameters. Given the identified parameters are often used as a basis for structural control and health monitoring, it is important to know the statistics of those estimates. Probability logic with Bayesian updating provides a rigorous framework to obtain VAR model coefficients, quantify their uncertainty and moreover, calculate the statistics of modal parameters derived from the VAR model. In this study, an approach based on the VAR and Bayesian inference is investigated to obtain the most probable value and statistical features of modal frequencies of a steel plate girder bridge.