Artificial neural network-based damage detection at element levels of tall buildings: A numerical investigation

Abstract

Detecting structural damage in tall buildings at the element level remains a significant challenge due to their complexity. This study extends the numerical investigation of an artificial neural network (ANN)-based approach to enhance the accuracy of damage detection in structural elements, especially columns. The research specifically focuses on a three-dimensional (3D) reinforced concrete (RC) high-rise building with 30 stories and a height of 90 m, assuming that the locations of damaged floors are known. To monitor vibrations, a single triaxial accelerometer is installed on each floor. Previous studies primarily focused on bending modes, making it difficult to accurately detect damage in columns. Therefore, this study incorporates both bending and torsional modes in the training process, as torsional effects naturally occur under real conditions. Additionally, the vertical components of the mode shapes are included to enhance the identification of column behavior, as columns in such buildings are typically loaded close to their capacity limits. Accurately capturing these vertical components is crucial for reliable damage detection. These modifications improve prediction accuracy. Particularly, 100% of damaged shear walls and columns are detected correctly, making the proposed method more effective in detecting structural damage, especially in columns.