Seismic-Induced Damage Localization Through Neural Networks and Clustring Method

For many essential structures, damage identification soon after earthquakes is of great importance because it can lead to a prompt decision about evacuation or occupation. In this study, a new algorithm is developed for localizing seismic-induced damage to structures. The proposed method only requires measured response accelerations and combines a sensor clustering technique with artificial neural networks (ANNs) to localize the damage. In this method, adjacent accelerometers are placed within a cluster and for each cluster, one artificial neural network is trained using the data obtained from the structure’s intact condition. The difference between the measured and predicted response accelerations by ANNs is calculated to define a Damage Sensitive Feature (DSF). The DSF is small when there is no damage in the cluster. However, when the structure is damaged during an earthquake, the DSF of the cluster will be larger than the intact condition. Therefore, the presence of damage in that cluster can be identified. The proposed method is applied to the Kuala Lumpur International Airport (KLIA) air traffic control (ATC) tower. A detailed finite element model of the ATC tower is developed and validated by comparing its natural frequencies with those measured on-site. For the training of neural networks, the 120 m tall tower is excited by 30 natural earthquake records, and its response accelerations are recorded at 11 locations to form 10 clusters. Results indicate that the proposed method efficiently localizes the seismic-induced damage to the tower under different earthquakes.