Damage Identification Research Of Structure By Means Of Wavelet Analysis And BP Neural Network

Modern civil engineering structures are becoming increasingly large and complicated with the rapid development of science and technology. Huge and complicated structures in the design life will be unavoidable to suffer the structural damage accumulation and decrease the structural resistance by the environment and structural factors, such as environmental erosion, natural disasters, long-term load effects, fatigue effects and material aging etc. There is a great safety risk in the structural service period. Therefore, the damage detection techniques have attracted more and more attention in the field of civil engineering. In this paper, a review for the damage identification approaches is given. In addition, based on the wavelet analysis and BP neural network, two-step damage identification method has been proposed. The validity and effectiveness of the proposed damage identification method has been verified by the numerical simulation examples.In the case of multi-damage identification by using the BP neural network, a problem is that the neural network training needs very large sample number. Thus, it is difficult to the practical engineering problems. To solve this problem, a two-step damage identification method is developed in this work by the combination of the wavelet analysis and artificial neural network. First, the wavelet analysis is applied to the damage location identification. Then, the neural network method is used to calculate the damage degree. Two-step damage identification method is applied to a single layer frame. The damage locations for four different damage cases are determined from the continuous wavelet transformation. The effects of the different type of modes, different order of modes and different finite element model are discussed and compared. Then, the damage degree is accurately calculated by applying the BP neural network. The results show that the two-step method truly has a high efficiency with the less training samples. The proposed method can be applied to the actual engineering structures.In addition, a three-span continuous beam structure and a four-story frame structure are also investigated in this work by using two-step damage identification method. The results of the three-span continuous beam structure are satisfied. However, for the four-story frame structure, the damage identification result is disturbed by the discontinuity of the nodal number in the FEM model. This problem can be solved by using the layered FEM model. That is the FEM model is divided by layers. Finally, the results are satisfied.