ABSTRACT
ABSTRACT: Civil structures degrade since their construction, which leads to inherent uncertainties and nonlinearities that cannot be modeled accurately. Detecting anomalies in the structural properties that may occur in such structures appears thus as a daunting task using conventional analysis tools. Artificial neural networks (ANN) offer a fair alternative to deal with such uncertainties and nonlinearities. Therefore, an ANN-based approach using measured dynamic responses is proposed to model the structure and, detect, locate and evaluate eventual damages that may occur in the structure. The technique relies on the direct use of vibration data measured on the undamaged system to train a network predicting the restoring force of each member, which models the structure and provides a reference for damage detection purpose. The damage parameter is assumed to be the time histories of the restoring force and corresponding stiffness. The network, fed with data of the structure subject to different loading and stiffness degradation cases, reconstructs then the actual restoring force loop and extracts the corresponding stiffness. Comparison of the reconstructed loops with the original ones allows damages to be detected as well as their location and extent to be evaluated. Application of the proposedANN scheme on a representative seismically isolated bridge structure is shown to provide efficient and accurate diagnosis of its structural changes. The proposed technique is believed to support further development in structural health monitoring of civil structures.
