In this study, a real-scale seven-story reinforced concrete (RC) shear wall structure is considered. The system identification is applied using the frequency domain decomposition (FDD) technique in the ambient condition of the building at various damage levels.
The building was tested in October 2005 and January 2006 at the University of California at San Diego. Four historical ground motions are used to vibrate the structure, and the progressive damage in the RC shear wall is monitored. The ambient vibration and White Gaussian Noise is applied after each motion record to find the updated modal properties of the building. To extract the modal properties of the building, the FDD approach is used in this study. Then, the finite element (FE) model is developed in the M-FEM program, and it is updated by a physics-based method (using the matrix update method) to match the identified modal properties at the undamaged state. The updated FE model is considered for damage detection. Two different techniques are applied for damage detection include of (1) a hybrid method in a data-driven approach (supervised machine learning) using a neural network (NN) (NN) a (2) physics-based method using the matrix update method.
Results show the outputs are well correlated in both methods, and the results are matched with observed damage in low and intermediate seismic level stages. On the other hand, the output shows a close correlation between the two methods and is mostly compatible with observed damage.
This research shows the efficiency of the developed hybrid method in case of limited access to the FE model.
