Analysis On Seismic Fragility Of Double-layer Curved Girder Bridge Based On Machine Learning

In this paper,a seismic fragility study is carried out using machine learning method to address the characteristics of double-deck curved girder bridges.The details of the study are as follows.1)This paper briefly introduced the general situation of machine learning and several machine learning technologies:linear regression,logistic lasso regression(LLR)and artificial neural network(ANN).The nonlinear mapping of LLR and ANN are applicable to most nonlinear problems.At the same time,lasso theory was introduced into the classical logistic regression,which can make the applied machine learning model to filter variables and reduce the deviation of model prediction.2)The inherent dynamic characteristics and seismic response of double-deck curved girder bridge were studyed.The response characteristics of brarings and piers in the double-deck curved beam bridge were obtained as follows:? The displacement of upper bearings is larger than that of lower bearings.With intensity of ground motion increasing,displacement deviations of the upper and lower layers shows a trend of decreasing first and then increasing.? When PGA is less than 0.2g,the whole pier is in the stage of linear elasticity.? The damage degree of the middle of piers is smaller than that of the top and bottom of piers3)The basic principles of the traditional single-parameter probabilistic seismic demand model for a double-decker curved girder bridge are reviewed.A probabilistic earthquake demand model is built using the machine learning method of ANN,which is compared with the traditional demand model and found that the ANN method can be Significantly improve the efficiency of seismic demand estimation.4)Multi-dimensional random seismic fragility analysis models of bearings,shear bolts and piers of double-deck curved beam bridge were established by using the proposed method.The influences of important parameters were analyzed,it was found that:? The gap between shear bolts is the most sensitive parameter in the limit state of slight damage and medium damage,even higher than the influence of ground motion intensity.? At the same time,the larger the gap between shear bolts,the lower the failure probability of piers.?With damage levels increasing,influences of the static friction coefficient and the concrete strength of bearings becomes smaller and smaller.?When serious damage was achieved,the seismic fragility of piers is mainly affected by intensity of ground motion.5)The traditional single parameter seismic fragility analysis model of double-deck curved beam bridge was established,compared with the multi-dimensional seismic fragility model by using the proposed method.Compared two fragility curves,it was found that:?The slope of the curve obtained by the proposed method is small and has a small dispersion.?With the limit state becoming higher,the median deviation decreases.At the same time,the advantages of proposed method are as follows:? The demand model based on ANN can avoid a lot of finite element simulations.?The probabilistic seismic demand model of double-deck curved girder bridge based on ANN method has higher reliability.?The regression coefficient of LLR regression model can identify sensitivity of parameters,and lasso method can automatically identify and reduce impacts of insensitive parameters.