Damage detection and modal identification of structural systems using sensor data

A promising application of Wireless Sensor Networks (WSNs) is in Structural Health Monitoring (SHM), where damage in structural systems can be detected using sensor data, avoiding or complementing costly and sometimes inefficient methods such as visual inspection. This thesis presents two effective SHM methods that use a structure's vibration response for local damage detection, and system identification for evaluating the global condition.;For local damage detection, influence coefficients, obtained from linear regression analysis between two sensor node responses are used as damage indices for determining the existence of damage. The accuracy of these coefficients is confirmed by considering two indicators: Evaluation Accuracy (EA) and Normalized Estimation Error (gamma). The change point of time-variant influence coefficients is determined using a Bayesian statistic. The effectiveness of influence coefficients as a damage indicator is demonstrated through laboratory experiments performed on a simple beam-column connection that represents a local structural joint subjected to different excitations. These experimental results, combined with the investigation of accuracy evaluation parameters, support the efficiency and reliability of the dense sensor network system in detecting damage in a local scale.;Two types of system identification methods, Eigensystem Realization Algorithm (ERA) and Auto-Regressive Moving-Average with eXogenous terms (ARMAX) models are examined for global damage detection. Modifications to these methods enables them to work in output-only systems presenting ERA with Observer Kalman Filter Identification (OKID), ERA with Natural Excitation Technique (NExT), and Auto-Regressive (AR) identification methods. A modified version of ERA-NExT is introduces to enhance the performance of the method. Numerical simulations are performed for a moment frame structure with Gaussian white noise ground excitation. Stabilization diagrams are plotted and computational costs for all methods are approximated. The identified modal parameters are compared with the reference structure and the accuracy of each method is evaluated. The numerical simulations show that the proposed system identification method is reliable and cost effective.