Application Of Recursive Plot And Convolutional Neural Network In The Field Of Bridge Damage Identification

The service requirements of long-span bridges are usually 100 years.In order to make the bridges meet the service life requirements,many scholars have carried out a large number of researches on bridge damage identification,and applied various signal analysis techniques to judge the damage status of bridges.However,the non-stationary nature of bridge signals and the complexity of environmental factors restrict the further development of traditional methods in this field Therefore,it is increasingly important to find new non-stationarity analysis schemes and feature mining techniques.This paper verifies the possibility of applying the method based on recursive plot and convolutional neural network in the field through damage simulation and simple supported beam test research of cable-stayed bridge,and quantifies the non-stationary signal changes through recursive quantitative analysis to characterize the change in damage degree trend.The specific research work is as follows:(1)First of all,this paper discusses the domestic and foreign research status of bridge damage identification,recursive analysis technology and convolutional neural network,and describes and analyzes the theoretical background,development status and research status of recursive analysis and convolutional neural network in detail.The basic theory of recursive technology was initially verified,and the technical and theoretical points required for recursion were mastered,which laid the technical and theoretical foundation for subsequent research.The recursive plot is the main research object of this paper.It transforms the one-dimensional acceleration signal into two-dimensional image information through phase space reconstruction,and mines the damage information contained in the acceleration signal.The convolutional neural network is used to classify the recursive plots,and the recursive plots representing different damage information are classified.Recursive quantitative analysis is the microscopic representation of recursive plots.It analyzes and calculates the information of recursive plots,and quantifies and qualitatively recursive plots.It is used to reveal changes in the degree of damage in this paper.(2)The above recursive plot and convolutional neural network are used for damage simulation of cable-stayed bridges,to explore how to use the two together for bridge damage identification,and to verify the possibility of applying this method in the field of bridge damage identification.Establish multiple sets of cable-stayed damage conditions,and perform recursive plot analysis and convolutional neural network classification on cable-stayed damage signals.The classification results show that the method has high accuracy,but the sensitivity to damage needs to be further optimized.(3)Use recursive periodic density entropy to mine the damage degree information contained in the signal.Considering the non-stationary characteristics of bridge damage signals,this paper uses recursive quantitative density analysis method to distinguish the non-stationary degree of different signals to characterize the change of signal damage degree.Through simulation,it is found that the recursive periodic density entropy can effectively reflect the non-stationary change of the signal,evaluate the damage degree change information contained in the signal,and improve the objectivity and effectiveness of the recursive plot to characterize the signal damage degree and damage location.(4)Finally,on the basis of recursive plots and convolutional neural network methods,this paper introduces the theory of recursive periodic density entropy and conducts a simple supported beam verification test.The possibility of applying this scheme in practice was verified through experiments.Through the recursive periodic density entropy analysis,the non-stationary change mode of the signal when the damage degree changed was obtained,which effectively improved the accuracy of identifying the damage degree.