It is very important to detect structural damage by using measured data as little as possible, especially in complex structures. A neural-networks-based method to diagnose damage by using the structural local modal strain changes is proposed in this paper. Presenting somewhere in the structure, damage will result in global modal strain changes, therefore, it is likely to obtain structural damage information by monitoring the changes of modal strain at specified locations which are sensitive to structural damage. In this study, firstly the changes of modal strains at specified locations in the structure in various simulated damaged cases by finite element method (FEM) are obtained. Second, a neural network is presented to establish a mapping of changes in local strain mode shapes and structural damage, and then neural network is trained using simulated data. Finally, the damage prediction ability of trained neural network is verified. The numerical results demonstrate that the proposed method is prospective in application.
