Research On Fatigue Damage Identification Of Stay Cables Based On GAF-CNN Method

As the main stress component of cable-stayed bridge,the fatigue problem of stay cables has been concerned by scholars at home and abroad.The research on fatigue damage identification of stay cables can provide guidance and reference for the daily maintenance and management of cable-stayed bridges,effectively prolong the service life of cable-stayed bridges,and has very important engineering application value.In this thesis,the vulnerability analysis of stay cables is firstly carried out,and the stay cables with the most fatigue damage are determined.Secondly,the refined finite element model of the stay cable is established,and the fatigue damage analysis is carried out.Finally,a method based on Gramian Angular Field and Convolutional Neural Networks(GAF-CNN)is proposed for cable damage identification.The main research contents include :(1)According to the theory of fatigue analysis,the evaluation index of cable fatigue vulnerability is established.The Monte-Carlo method is used to establish the random train load model and the cable fatigue load spectrum is obtained by the rain-flow counting method.Through the vulnerability analysis,the most vulnerable cable is found.(2)Taking the parallel steel strand stay cable as research object,considering the uneven force of each steel strand,a refined finite element model of the stay cable is established,and the fatigue analysis is carried out to obtain the fatigue life of the stay cable.The results show that the fatigue life of the stay cable in the anchorage zone is shorter,and the outer steel wire is more prone to fatigue damage than the central steel wire.(3)Based on the refined model of the stay cable,the acceleration response time history is obtained,and the acceleration signals are converted into a two-dimensional image by using the Gramian Angular Field(GAF)to construct the GAF data set.The Res Net34 network suitable for cable damage identification is constructed by using transfer learning.The constructed GAF data set is input into the network model for training and testing,and the damage degree and damage location of the cable are identified,and the ideal recognition accuracy is obtained.On this basis,the influence of sensor position and noise on the recognition results is discussed.