Damage identification for high-rise buildings using an eigen problem based approach and an artificial neural network

Structural health monitoring (SHM) has been applied, albeit in small numbers, in the regular control of high-rise buildings' health that has deteriorated having been subjected to any loading. The main objective is to detect invisible damage at the initial levels for early warning of collapse and retrofitting works. However, in practice, it is obvious that monitoring is preferred preferred for general purposes instead of these levels for many different reasons, and it is implemented utilizing dense sensor networks and high-qualified sensors in very few examples. To circumvent the complexity of high-rise buildings, hybrid procedures are proposed to detect damage at storey and then element levels. For the first step, three approaches, namely modal strain energy (MSE), eigenvalue problem-based inverse solution, and Artificial Neural Network (ANN) are considered. The inverse solution based technique is constructed based on the behavior of lumped-mass systems simplified from full buildings employing the Guyan static condensation procedure. The technique detects damage accounting for only the first two lowest bending modes and is verified numerically and experimentally. The ANN approach is also implemented for storey-level detection. The networks are trained using only the lowest bending modes. In the second step, the ANN method is deployed to pinpoint damaged elements focusing only on the identified storeys, effectively reducing the number of variables. In this thesis, hybrid approaches are implemented considering the two-dimensional (2D) and then (3D) 30-storey buildings. As a result, storey-level and element-level detection is accurately achieved as long as the generated modal data is low-level noise-contaminated.