Identification Of Damaged Stay Cable Of Long-span Bridge Based On Machine Learning

The research object of this paper is a cable-stayed bridge(the prototype of which is Suramadu Bridge)and the health monitoring methods of the main component cable are studied.First,on the basis of the finite element model of the bridge,the working conditions were determined according to different damage levels and the number of damaged cables,and the strain data of the pull-down cables under different working conditions were obtained.Then,based on the BP network,SCG(Scaled Conjugate Gradient)network and RBF(Radial Basis Function)network,the damage identification methods of long-span cable-stayed bridges are studied.Faced with many shortcomings of traditional BP neural networks,such as slow algorithm convergence rate,low learning efficiency,difficulty to determine the network structure,easy to fall into the local minimum point,etc.By adding momentum items and adaptive learning rate,an improved BP neural network can be obtained.Using the "Cross-validation method" to investigate the influence of different parameters on the performance of neural networks and select the proper network in the end.The research shows that the improved BP neural network is able to identify the damage of the cable within a limited range of accuracy.On the basis of conjugate gradient algorithm,"SCG neural network" is selected as a damage identification method for cable structure.By establishing the same network structure as the BP algorithm and statistically comparing the performance of the network on the same validation data set,the advantage of "SCG neural network" compared to BP neural network is proved.To sum up,the "SCG neural network" can be used in a large-scale network structure,and the result shows that the "SCG neural network" is able to obtain a better damage identification effect of the cable than the BP network.The RBF neural network is used to identify the damage of the cable and the identification results of the network on the data samples of the same validation set are statistically analyzed.Finally,the advantages and disadvantages of the RBF the SCG neural network are analyzed.The research shows that the damage recognition effect will only decrease when the damage reaches a certain level.From a whole point of view,the RBF neural network can effectively and stably recognize the damage of the cable.