Seismic Analysis Method Of Steel Box Arch Bridge Based On Artificial Neural Network

Steel box arch bridge is widely used because of its strong spanning capacity and high bearing capacity.Because seismic loads are uncertain,variability in propagation and extremely expensive in computing time.This makes it the most important factor restricting the development of steel box arch bridge to large span.In order to solve the problems that have puzzled the seismic design of long-span steel box arch bridges for a long time,this paper analyses the correlation of various factors,and uses artificial neural network and some signal analysis methods to solve them.Taking Datengxia approach channel bridge in Yunnan province and Rokko island bridge as well as Nishinomiya port bridge damaged by the Kobe earthquake as the background bridge.By means of finite element model and shaking table test,based on seismic vulnerability curve and seismic damage probability,the solution proposed in this paper is verified.It is found that the efficiency,accuracy and stability of this method are greatly improved compared with the traditional method.Specific research results include:(1)In view of the current artificial ground motion generation model,only one-sided attention is paid to the fitting of acceleration response spectrum,while ignoring the respect for the true nature of ground motion.Based on the analysis of the principles of existing artificial ground motion generation models,this paper proposes to improve the Ghodrati model by using artificial neural network and multiple empirical mode decomposition(MEMD).The new model uses artificial neural network to learn the relationship between acceleration response spectrum and ground motion amplitude and frequency information.Compared with the traditional model,this model can minimize the serious deviation from the nature of natural ground motion under the premise of guaranteeing the fitting accuracy of acceleration response spectrum.Combined with Datengxia approach bridge,Rokko island bridge and Nishinomiya port bridge,based on a variety of seismic samples,by means of finite element model and shaking table test,the time-frequency acceleration response spectrum,seismic vulnerability curve and seismic damage probability are analyzed.It is found that compared with the traditional artificial ground motion model,the proposed model can better reflect the nature of’the original ground motion.(2)In view of the severe variation of ground motion in the near-fault area of earthquake,the commonly used spatial variation model of ground motion is difficult to use in engineering.Based on the analysis of the principles of various spatial variability simulation methods.this paper proposes a spatial variability model based on multiple ground motions using artificial neural networks.This model constructs the relationship between ground motions and coordinates in artificial neural network,and uses this relationship to simulate spatial variability.Combining with the approach bridge of Datengxia bridge,Rokko island bridge and Nishinomiya port bridge,the spatial variability recorded in the near-fault region is compared based on several groups of near-fault ground motion samples.It is found that compared with the traditional model,the model in this paper has very high simulation accuracy of spatial variability,and can reduce the seismic response and probability error of seismic damage caused by spatial variability error.(3)The large amount of seismic calculation of bridges restricts the use of long-span bridges.Based on the analysis of the present situation and existing problems of structural damage prediction using artificial neural networks,this paper presents a model for predicting seismic damage of bridges using artificial neural networks.Based on a large number of training samples,this model constructs the relationship between sample parameters and damage index in artificial neural network,and uses this relationship to predict the seismic damage of unknown samples.Combining with Datengxia approach channel bridge,the model is optimized.It is found that the model can realize rapid evaluation of seismic damage under the condition of guaranteeing accuracy.And with the increase of training samples,the prediction effect can be further improved.