The damage detection system of a real steel truss bridge was developed to identify the location and severity of the damaged members. At first, the loading test was performed to characterize the real bridge. The real steel truss bridge was measured by electrical strain gages and accelerometers when the train passed. The measured strains and acceleration were used to refine the stiffness and the mass of the finite element model. The damage scenario, that can be happened in the real situation, was simulated by the refined finite element model. The damage localization was implemented to classify the damaged part in the bridge by the neural networks. The neural network was constructed as two steps: at 1st step, the half-span, which had some damages occurred, was found, and at 2nd step, the severest abnormal part in the total 8 parts of the real bridge was detected. The learned neural network was verified by the used data.
展开▼
