Damage Detection Of The Structures Using Neural Networks With The Srain Mode Shape

The structural damage detection and diagnosis has become one of the advancing fronts of the civil engineering researches, for more and more the existing structures need to be surveyed and strengthen with the yield designed date coming. The dynamic characteristics of structures are in closed relationship with the structural parameters, and structural damage will cause the dynamic characteristics shifts correspondently. So the detection and diagnosis of the damage structures can be realized used the dynamic measurements if the mapping relationship between the damage and the parameters had been founded. The neural network technique has shown great superiority in structural damage detection and diagnosis for its strong non-linear mapping ability, rapid computation and anti-interference capability. But there are still some problems need to solve, for example, the selection of the neural networks model, the input parameters and so on. The cantilever beams are selected as the representative structure to study in the paper. Radial Basic Function neural networks are employed to diagnose the crack, which is simulated by splits created by the saw, for the strain mode shape data are choose as the input parameters. Different neural networks, one or both the frequency and the strain mode shape is selected as the input parameters, are compared that proved the strain parameters are better than the frequencies' as the input parameters. The development of the neural networks was applied to the mechanics is reviewed, and the feasibility of the neural networks used in the fragment mechanics is studied. Here, a example of the J-integral is calculated with the neural networks method. All of above studies have founded a credible basic for the future research.