Research Of Artificial Neural Network In The Damage Detection Of Frame Structure

At present, structural damage diagnosis has become one of the advanced topics of civil engineering researches. The structural damage will leads to some changes in inherent frequency and vibration modal. So, if the mapping relationship between dynamic character change and structural damage can be established, the damage can be identified using vibration test information of the structures. The artificial neural network(ANN) as a new method has been developed rapidly and used widely. Because its high effective ability of nonlinear mapping, powerful ability of solving contrary problem, ability of compute in real-time, its an excellent method in the field of damage identification, In this paper, The feasibility and availability of the proposed method are theoretically proved in a numerical case by using an five storey symmetric framework. The main research content includes:1. Structural damage detection is divided into three sub-modules base on inherent frequency and vibration modal, which are structural damage identification module, structural damage localization module and structural damage severity determination module. For each step, different modal parameters( NFCR_i,FFC_i and NDΦ) which point damage feature are inputted to different neural network (PNN,RBF and BP) as parameters for the stepwise structural health monitoring.2. As far as structure damage is concerned, the inherent frequency and vibration modal is very sensitive at different stage. Using different compages of structural modal parameters (Modal Frequency, Model Shape) before and after damage, a new damages index NDΦbased on the model shapes has been proposed. And through the mathematics derivation related the relationship between indexs and damage location or degree has been carried out. Reasonable input which contains more damage information has been chosen.3. In study, the structure model of a five storey framework is built and analysed by ansys software in order to get the modal data. Then the inherent frequency and vibration modal is computed. The result is the input of the ANN. The samples of training and testing are obtained by reducing stiffness to simulate the degrees of damage. Although the identifying effectiveness of damage locations is not better than the results of theoretical analysis, they are correctly located. So the proposed method in the paper is practical.