Dynamical Diagnosis And Research On Frame Structures Based On Modal Datas And Neural Network

Damage identification and health monitoring of the structures have become an international research focus in recent years,and so is the artificial neural network which is used in all areas of civil engineering and is especially significant in structural damage detection.Based on the experimental modal data and artificial neural network method,the sway brace in the framework structure is identified in this paper.Firstly, the sway brace was set for the frame structure model in the laboratory, taking the method of setting steel supports in different locations.Secondly, mode tests were carried out on this frame structure of different sway brace locations to get the initial state data.Not only the translational mode was got in the test, but the torsional mode was also got.An index was proposed to determine which floor the sway brace is in,based on the modal data of torsional mode shapes.However, the index could only be used in the identifycation of the asymmetrical sway brace.Then we used neural network method to identify symmetrical and asymmetrical supports.Moreover, the data of translational mode and torsion mode were used to be an input vector for identifying the frame structure more accurately. Thus this paper proposed a new combination index mark for working as the input vector. In order to make the training samples of neural network more real,firstly took the experimental data of the initial state of the frame structure to update the model.Then analyzed the model by SAP2000 software,found that it was consistent with the experimental data. After updating the model,this paper carried out the experiment of identifying the position and section stiffness of steel support based on the neural network multi-step identification method.This method distinguished the identification of layer locations from that of the section stiffness.First we made sure the layer position of the support, and then defined the specific position of support inside the layer which is known. Later we carried out the identification of section stiffness of the support.Thus could reduce the study sample by taking this method and improve the study efficiency. Here used the PNN neural network to determine the position and RBF neural network identify the section stiffness of steel supports.At last, analyzed the identified model by SAP 2000,it was found that the result of the numerical simulation and the result of the experiment was influenced.