Performance Comparison of Different Autoregressive Damage Features Using Acceleration Measurements from a Truss Bridge

Abstract

Time series analysis has been applied to structural monitoring signals for system damage identification in a number of research literatures. Among various time series analysis tools, univariate autoregressive modeling (AR) is one of the most commonly used methods because of its innate computational efficiency. In this paper, three autoregressive damage features extracted directly from the ambient vibration data and from the vibration signal autocorrelation will be presented. Two of the features are distance functions of AR model parameters and the third feature is a function of AR residuals. These features are then applied to acceleration measurements collected from a member of a truss bridge to detect a structural change, and their performances are compared and commented.