Structural damage identification of high-rise buildings: An artificial neural network based hybrid procedure

Abstract

Structural damage detection of high-rise buildings is by far not reached because of their complexity. In this study, an artificial neural network (ANN) method-based two-step approach is suggested to detect damage at element levels of a 3D 30-storey 90 m high RC building containing 2880 degrees of freedom (DOFs). One biaxial accelerometer per floor is erected, making the number of measured DOFs equal to about 2% of the full system. Only the first three bending modes in the orthogonal axes are accounted for. A network is constructed in Step 1 to detect damaged storeys based on the similarities between tall buildings and beam-like systems. All components' stiffness parameters of each storey are assigned to one variable. In Step 2, another network is built focusing only on the detected storeys to localize ruined elements. Furthermore, aiming at detecting damage considering modal data generated under ambient conditions, inevitable measurement noise effects are also considered to challenge the proposed ANN technique. As a result, the light and robust networks lead to precise storey- and element-level detection promptly as long as the desired vibration-based properties are free of noise as well as noise-corrupted.