Numerical And Experimental Studies Of Joint Damage Identification For Frame Structures

At present,many existing engineering structures have been in the old age,and they are to be diagnosed,maintained and strengthened,so structural damage diagnosis has become one of the advanced topics of civil engineering researches.The dynamic characteristics of structures are in close relationship with structural parameters,and structural damage will cause dynamic characteristics change correspondently.If the mapping relationship between dynamic characteristics change and structural damage can be established,the damage can be identified using vibration test information of the structures.The neural network technique has shown great superiority in the structural damage diagnosis field for its strong non-linear mapping ability and anti-interference capability.But there are still some problems to be sovled such as the selection of neural networks algorithm,determination of networks input vector and noise error of test data,and they are the main factors of identification accurace and networks training velocity.In the recent years,back-propagation neural networks(BPNN) is the most used widely neural networks.But traditional BPNN has weakness itself.Based on predecessors' studies, BPNN is improved in two aspects in this paper.As a result, the performance of traditional BPNN has been made better in the extent.In this paper,some joint damages of widely used frame structures are dicussed, and a joint damage identifying approach using improved BPNN in which combined parameters are chosen as input vecter for frame structures is presented joint damages are diagnosed through two steps,and the noise injection training techniche is introduced to enhance the anti-interference capability of the neural networks. Finally, it is proved that improved BPNN has stronger anti-interference capability than traditional BPNN.The feasibility and availability of the proposed methed are theoretically proved in a numerical case by using an four-storey symmetric steel frame.However,further research should be done to improve the precision of damage extent detection. Furthermore,experimental modal analysis of a two-storey steel frame is too analyzed.the change results of modal frequency from experience are accordance with theorical analysis.Although the identifying effectiveness of joint damage locations is not better than the results of finite element analysis,they are correctly located.so theproposed method in the paper is practical.