Seismic damage identification in buildings using neural networks and modal data

A seismic damage identification method intended for buildings with steel moment-frame structure is presented in this paper. The method has a statistical approach and is based on artificial neural networks and modal variables. It consists of two main stages. The initial one is devoted to the calibration of the undamaged structure and the final one to the identification of the damaged structure after an earthquake. The inputs of the nets are the first flexural modes (frequencies and mode shapes) at each principal direction of the structure and the outputs are the spatial variables (mass and stiffness). A damage index at each storey is determined by comparing the initial and final stiffness. A simplified finite element model was used to generate the data needed to train the nets. This model is consistent with available modal data and damage definition. The method was simulated on a 5-storey office building under conditions as close as possible to reality. The robustness of the method was verified with simulated data. Latter on, a sensitivity analysis of the mass variability was also carried out. Finally, the influence of modal error in the accuracy of damage predictions was statistically studied. Results are successful as concern as the robustness of the method. However, it is found that this approach is quite sensitive to modal errors.