Damage Identification Of Cable-Stayed Bridges Based On Optimized Intelligent Algorithms And Low-Order Curvature Modes

With the rapid development of engineering technology,the number of long-span bridges across the cutting is increasing.Among them,cable-stayed bridge's panning capacity ranks in the front because of its high efficiency,light weight and clear concise transmission path.Because of its beautiful shape,it not only takes the role of key traffic roads,but also often becomes the landmark structure of the region.Therefore,health monitoring of cable-stayed bridges is of great significance,and damage identification is one of the key links in health monitoring system.With the support of the National Natural Science Foundation of China(No.51778088),a more efficient and universal damage identification method for cable-stayed bridges is studied from dynamic parameters and intelligent algorithm.In order to verify the feasibility and accuracy of the identification method proposed in this paper,the damage identification for Yamen Bridge with double pylons and single cable plane pre-stressed concrete box girder system is carried out.The main work and conclusions are as follows:(1)The theoretical derivation and parameter characteristics of each dynamic parameter are compared based on practical application,and the low-order curvature mode is chosen as the dynamic parameter for damage identification.1)The dynamic parameters measured in engineering are reliable in the low-order part,and the high-order dynamic parameters often bring more noise effect.The principle of damage identification in this paper is to use only the low-order dynamic parameters.2)The theoretical deduction process of each dynamic parameter is analysed.Noting that the curvature mode didn't directly deduced by using the parameters reflecting the overall dynamic characteristics of the structure,therefore a method for identifying multiple damages by dividing them into single damages is proposed.For statically determinate simply supported beams,the curvature modes of first-order vertical mode is subjected to second-order differential operation(the numerical realization is second-order difference).According to the results of second-order differential with large damage change,the influence elements of multiple damages are separated and the curvature modes of multiple single damage are calculated,which are used as input data for damage identification.For multi-time statically indeterminate cable-stayed bridges,the curvature mode under non-invasive condition is subtracted from the lower-order curvature mode under multi-damage condition,and the multi-damage influence elements are separated according to the larger curvature modal difference of damage change value.The curvature mode under multiple single-damage conditions are obtained as input data for damage identification.3)Due to the limitation of the algorithm when using curvature mode to divide multi-damage into single-damage,the parameter will lose some structural information,and the identification performance of curvature mode with noise is not ideal.Therefore,this paper will use optimized intelligent algorithm to identify damage with curvature mode parameters to achieve ideal anti-noise effect.(2)The essence of structural damage location identification is pattern recognition,and the essence of damage degree identification is the expression of non-linear mapping.By comparing and analysing the characteristics of intelligent algorithms,it is noticed that the neural network based on Radial Basis Function(RBF)has a clear mathematical idea of pattern clustering,and it can implement interpolation of complex functions,which meets the needs of bridge structural damage identification.In this paper,RBF neural network is used as the basic structure of intelligent algorithm,and the following optimization is carried out:1)The input layer-hidden layer of RBF neural network is the unsupervised learning stage to implement pattern clustering.The traditional algorithm calculates and determines the clustering center and the clustering radius by itself.In this paper,the dynamic parameter samples with 50% damage of each unit are designated as the clustering center before the network training,which improves the convergence speed of the algorithm training.2)The hidden layer-output layer of RBF neural network is the supervised learning stage of error back propagation.In order to avoid falling into the local minimum of network training caused by random initial weight matrix,the idea of global minimum is obtained by extensive search in solution space using genetic algorithm.A program is designed to optimize the initial weight matrix.(3)In order to verify the validity of low-order curvature modal dynamic parameters as input data for damage identification,and to confirm the feasibility and accuracy of the optimized intelligent algorithm,a simple-supported beam bridge example is used for damage identification research.1)The optimization intelligent algorithm is compiled by MATLAB language and trained,and the convergence speed of network training before and after the optimization of genetic algorithm is compared,which proves the effectiveness of genetic algorithm in improving the convergence speed of network.2)Considering nine single damage and three multi-damage conditions,the identification effect of the optimized intelligent algorithm is tested.The first-order curvature mode is successfully used to accurately identify the location and degree of damage under single damage and multi-damage conditions.3)Considering different noise levels,white Gaussian noise is added to the curvature mode,and the anti-noise performance of the optimized intelligent algorithm for identifying damage location and degree is studied on the basis of the simply supported beam bridge model,which proves the excellent anti-noise ability of the algorithm.(4)Damage identification of numerical model of Yamen Bridge with double pylons and single cable plane pre-stressed concrete box girder system is studied.Damage identification of beam and cable is carried out by using the algorithm respectively.1)The first-order curvature modal difference dynamic parameters of the structure and the corresponding optimization intelligent algorithm which is programmed by MATLAB language are used to identify damage of beam and cable respectively.2)For the damage identification of main girder,10 single damage and 5 multi-damage conditions are considered,and the trained algorithm is used to identify the damage location and degree of beam of cable-stayed bridge accurately.For the damage identification of cables,4 single damage and 7 multi-damage conditions are considered,and the results show that the optimized intelligent algorithm can accurately identify the damage location and degree of cable damage.3)Considering various noise levels,taking the beam and cable of Yamen Bridge as research object,the anti-noise performance of the optimized intelligent algorithm for damage location and degree identification is studied,which proves the high-quality anti-noise ability of the algorithm.In this paper,a method of damage identification for cable-stayed bridge structures is proposed.Low-order curvature modal can not only reduce the influence of noise in data source,but also can divide multiple damage nto multiple single damage for identification.An intelligent damage identification algorithm based on RBF neural network is designed,and the problem adaptability of the clustering center of the algorithm is adjusted.Learning from the idea of genetic algorithm,the weight matrix of RBF neural network is optimized,the results show that training convergence speed and mapping learning ability of the algorithm is improved.The damage identification of the beam and cable of Yamen Bridge is studied.The feasibility and accuracy of the damage identification method for cable-stayed bridge with numerical model based on low-order curvature mode and optimization intelligent algorithm are verified.The research results validate the feasibility and accuracy of damage identification method for cable-stayed bridges based on low-order curvature mode and optimized intelligent algorithm,as well as the good anti-noise performance of the algorithm.