Research On Bridge Damage Identification Method Based On Modal Analysis Theory And Neural Networks

With the expeditious development of our civil traffic projects, plenty of oversize bridges have constantly rushed, the number of new and old bridges has been increased. In order to insure safety and security of the people's health and wealth, it is the focus of current bridge project to detect fleetly and effectively structural damaged position and extent likely occurring and to command the health status of bridge in commonly using state in time.On the basis of collection and analysis of the data about structural damage identification and artificial neural networks, combined with the prospect of bridge damage identification and artificial neural networks, vibration modal analysis theory is integrated with BP neural networks so as to detect bridge damage with the help of ANSYS and MATLAB, at the same time, it has been successful in detecting damaged position and extent likely occurring, the method based on vibration modal analysis theory and BP neural networks is presented originally in this paper.Firstly, the damage identification method based on vibration modal analysis theory is discussed and analyzed. During the process, natural principle and detected procedure of this analysis theory are introduced, natural principle,predominance and shortcomings of every capabilities and applicable scope about these methods based on modal frequency,modal vibration shape difference,curvature modal shape/ strain modal shape,curvature modal shape difference,flexibility difference and flexibility curvature are systematically analyzed. In addition, these methods are adopted to the calculation of damage imitation to one cantilever beam, the study and comparison about diagnose ability based on series of methods to the different conditions of single damage location and a couples of ones are carried out.A new method, called flexibility curvature method, to localize the damage in structures is presented in this part. The method is not only highly-sensitive to damage, but also able to identify the damage location in structures without baseline modal parameters. Furthermore, the method has advantages of involving a small quantity of calculation and being simple and easy in use. The damage in structurescan be detected with satisfactory precision by using only a few lower modes. Besides, a method was presented for identification of multiple damaged locations in structures using curvature mode shape and flexibility curvature.Secondly, the damage identification method based on BP neural networks is discussed and analyzed. Introduced the basic theories of artificial neural networks in brief in this process, introduced BP neural networks, BP classic algorithm and the Lavender-Marquardt algorithm on the condition of optimization, and BP neural networks that realized in the MATLAB in details, natural principle and detected procedure based on BP neural networks and neural networks tool box functions are systematically analyzed. On this foundation, one simple supported rectangle steel beam model is presented as calculated example for the above-mentioned method application.Being aimed at the main drawbacks of slowly learning convergent velocity and easily converging to local minimum of the Backward-propagation (BP) networks is optimized in this part. Later by applying software MATLAB emulation, good effects are gained by adopting the optimized BP networks to process damage detection on a rectangular beam.Finally, according to the design and construction data of highway cable-stayed bridge on Yunyang Changjiang river, damage-detection-oriented finite element model of the bridge is established and the free vibration analysis is then carried out. The focus of the research is placed on three instances, including: one component of the bridge is damaged; two components and three components are damaged. Modal frequency, displacement modal shape and curvature modal shape are used as BP neural networks import vector respectively; sample data of each damaged state are collected; nine BP neural networks models are established for researching bridge damage d...