Research On Damage Identification Of Long-Span Self-Anchored Suspension Bridge Based On Characteristic Index And Neural Network

The long-span self-anchored suspension bridge has greater flexibility,and it is prone to large vibration and damage under the action of dynamic load.Therefore,the research on damage identification is of great significance in its detection,maintenance,reinforcement and even reconstruction,aiming at the service period of this type of bridge.Thus,based on the research results of displacement influence line,wavelet packet analysis and neural network principle,a damage identification method of preliminary damage judgment,precise damage location and damage degree analysis is proposed in this paper.Moreover,the applicability of the method is verified by an example,and the feasibility analysis of the BP neural network based on the hybrid learning strategy is carried out.Further,taking the expressway long-span self-anchored suspension bridge as the research object,the applicability of the damage identification analysis method proposed in this paper to this bridge is demonstrated and studied.And,the influencing factors are analyzed for the identification method,including random traffic flow,variable speed vehicle load,noise disturbance,etc.The main research contents are:(1)Structural damage identification method based on displacement influence line and wavelet packet analysisAccording to the principle of displacement influence line,these two damage characteristic indicators are constructed,deviation of displacement influence line(DDIL)and deviation curvature of displacement influence line(DCDIL).These are used to preliminarily judge the damage status of the bridge structure and to roughly locate the damage area.According to the principle of wavelet packet analysis,the relative energy rate of wavelet packet energy spectrum(RES),this damage characteristic index is constructed.And it is used to precisely locate the damage location of bridge structures.Therefore,the effectiveness of the proposed method is verified by taking a simply supported girder bridge as an example.The results show that the comprehensive identification effect of the RES curve constructed with the displacement time-history signal is the best among all bridge dynamic response types;the identification results are less affected by factors such as vehicle speed,vehicle weight and bidirectional traffic.(2)Identification method of structural damage degree based on optimized BP neural networkBecause of these properties,the parallel computation of the Genetic Algorithm and the local probability escape of the Simulated Annealing Algorithm,a BP neural network based on a hybrid learning strategy is proposed,Genetic Simulated Annealing Algorithm Optimized BP Neural Network(GASA-BP).Taking a simply supported beam bridge as an example to identify the damage degree,the damage identification results of these neural networks are compared and analyzed,including BP,GA-BP(Genetic Algorithm Optimized BP Neural Network)and GASA-BP.The results show that GASA-BP effectively improves the convergence performance and fitting ability of the algorithm.And the overall identification error and error rate of structural damage are minimal.It can accurately predict the extent of damage.These prove that the optimized neural network is suitable for the identification of the damage degree of the structure.(3)Structural damage identification analysis of long-span self-anchored suspension bridgesThe finite element model of the long-span self-anchored suspension bridge is established in ANSYS software,including the use of the segmented catenary method to determine the reasonable alignment of the main cable,the consideration of the initial strain of the main cable,suspenders,main beams,bridge towers and other components to achieve a reasonable bridge state,and the employ of subspace iteration method to analyze the natural vibration characteristics of the structure.The structure of this bridge was subjected to damage identification analysis using the proposed damage identification analysis method.The structural damage was preliminarily judged by the indexes of DDIL and DCDIL.On the basis of the roughly located damage area,the RES index is used to accurately locate the structural damage.Finally,GASA-BP was used to identify the damage degree of the structural damage site.In addition,influencing factors are analyzed in the identification results,including random traffic flow,variable speed vehicle load,noise disturbance,etc.The results show that the method proposed in this paper is suitable for the analysis of the damage location and damage degree of one or more parts of the structure.Factors such as random traffic flow and variable-speed vehicle load have obvious effects on the magnitude of the RES damage index,but do not change the damage identification law.Signal noise has a significant impact on damage identification results.As the level of signal-to-noise ratio decreases,the recognition accuracy decreases in a nonlinear trend.