Application Of Artificial Neural Network And Wavelet Analysis To Damage Identification Of Girder Bridge

With the development of the modern complicated structures, structure health monitoring and damage identification are becoming more and more important. Artificial neural network and wavelet analysis have become two core techniques in structure damage identification. This thesis concentrates on the damage identification of girder by using the two techniques combined with multi-step method.Firstly, in this thesis, some researches have been done for load identification and ultimate load bearing capacity prediction of the prestressed concrete filled rectangular steel box girders based on real model test data.Secondly, putting all the signals into the neural network for one time would lead to some abnormal results, such as unconvergence, long training time, low training precision and so on, because there are too many signals which should be gathered from a complicated structure. To solve this problem, complicated structure should be decomposed to a series of substructures according to certain regulations, and then calculation model is set up for each substructure, which is treated as a unit. In this thesis, a damaged plane truss model has been studied by using neural network in two steps. In the first step, the truss is decomposed to three substructures and the damage could be located to a certain substructure by a trained neural network. Then the damage could be located to a certain bar and the damage degree could be identified by another network in the second step.