Detection, Localization and Quantification of Anomalies in Mass, Stiffness and Damping Based on Time Series Modelling Using Output-Only Data

摘要

In this paper, an improved time series approach for anomaly detection, localizationand quantification using output-only data is presented. It was shown in previousstudies conducted by the authors that this methodology was able to detect, locate andquantify the changes in the mass, damping and stiffness parameters in a numericalmodel for noise-free conditions accurately. In this paper, the approach is extended fordamage identification when artificial noise is added to the data. This approach ismainly based on the idea that the changes in the structural properties of aninfrastructure can be related to the change in free vibration response (displacement,velocity and acceleration) of the structure. In this approach, the sensors of a structureare first categorized into different clusters according to their locations. Then, ARXmodels (Auto-Regressive models with eXogenous input), where only output dataunder free vibration are used, are created for different sensor clusters. Building theseARX models for baseline and damaged structures, the location and degree of theanomaly can be identified on the basis of the changes in the coefficients in the models.In order to verify the approach, it is applied to a numerical structural modelrepresenting a laboratory specimen. The results show that the change of mass, stiffnessand damping for noisy conditions can be identified separately by using this approacheven if these changes occur simultaneously.