| The suspension bridge with super span ability,full play of material performance,beautiful appearance and reasonable structure has become the main bridge type of long-span bridge.However,under the coupling effect of natural complex environment and long-term variable load,the suspender,as one of the load-bearing components of suspension bridge,is prone to damage,which poses a huge threat to the safety and durability of the whole structure.It is of great significance for scientific research and engineering application to fully mine bridge monitoring information,accurately identify the location and degree of suspension bridge suspender damage,so as to ensure bridge safety and road network mobility.This paper takes the world’s largest self-anchored track suspension bridge-Chongqing Egongyan Self-anchored Suspension Bridge as the engineering support.And on the basis of reviewing of the research status of existing cable bearing bridges suspenders/cable damage identification methods and the application of deep learning in Structural Health Monitoring,this paper uses theoretical analysis,numerical simulation,computer programming and other research methods,combining Response Surface Method and Whale Optimization Algorithm,Deep Convolutional Neural Network and Long short-term Memory Neural Network,to carry out the research of suspender damage identification method of long-span track suspension bridge based on Deep Learning.The main research contents and conclusions of this paper are as follows:(1)The theoretical basis of Deep Learning model involved in this study is systematically introduced,and the network structure and algorithm theoretical derivation of Deep Convolutional Neural Network(DCNN)and Long Short-term Memory Neural Network(LSTM)are mainly studied,which provides theoretical guarantee for the follow-up suspender damage identification based on Deep Learning model.(2)Based on the establishment of the initial finite element model of Egongyan self-anchored suspension bridge,this paper puts forward the structural finite element model modification method based on the Response Surface Method&Whale Optimization Algorithm(RSM-WOA).By comparing with the measured data,the accuracy of the model modification method is verified,and the modified finite element model can be used as a benchmark model for suspender damage identification research.(3)According to the characteristics of the acceleration response of the structure which is easy to obtain,combined with deep learning theory,a method for damage identification of suspension bridge suspender based on Deep Convolutional Neural Network(DCNN)acting on time-domain acceleration signal is proposed.The network model can extract the damage features from the acceleration data directly for the intelligent identification of suspender damage,eliminating the artificial construction of sensitive damage indicators.The DCNN model is built by Python and Pytorch,and the acceleration datas of 15 different kinds of suspender damage conditions are obtained by using the modified model for DCNN model training and testing.The result shows that the accuracy of damage location based on DCNN model is 90.07%,and the absolute relative recognition error of damage degree prediction is 10.60%.The research shows that the damage identification method based on DCNN can realize the identification of the location and degree of the suspender damage of the track suspension bridge,which proves the feasibility of using DCNN to extract the damage characteristics from the acceleration data.(4)Aiming at the problem that the Deep Convolutional Neural Network is difficult to extract the time-domain features of the acceleration time-history response,an improved method for suspender damage identification based on the Deep Convolutional Neural Network and the Long Short-term Memory Network(DCNN-LSTM)is proposed.In this combined model,the time series feature of acceleration signal is extracted by LSTM,and the feature extracted by DCNN is fused to realize the intelligent identification of suspender damage.Through the simulation experiment of computer programming,model training and testing are performed using the same acceleration data sets,and compared with the results of the suspender damage identification method based on DCNN.The result shows that the accuracy of damage location of suspender based on improved model reaches 94.00%,and the absolute relative identification error of damage degree prediction is reduced to 8.00%.The research shows that the improved damage identification method based on DCNN-LSTM can make full use of the effective information contained in the acceleration data to improve the accuracy of identifying the location and degree of suspender damage in track suspension bridge. |
