Intelligent Algorithms-based Fragility Analysis Of Isolated RC-Continuous Highway Bridge Using LRB

Because analytical method is the only way to get fragility curves of bridge strctures,and so far the related vulnerability research of isolated bridges is still very rare, it’s helpfulto do research in this field with the method based on intelligent algorithms and get thefragility curves of the structure. As isolated continuous girder bridge is a classsic type ofhighway bridges, it will be useful to do research in the damage probability when this kindof bridge is subjected to earthquake, and more comprehensive and more objectiveevaluation of the reliability of the highway traffic system can be made. A typical isolatedRC highway bridge is chosen for study, performance-based demand and capacity analysisof bridge components are performed based on the probabilistic model, and seismicfragility curves of bridge components and overall bridge structure are obtained in the wayof cloud method, RBFNN-based method and SVM-based method. The outline of thedissertation is as follows:1. Finite element analysis software SAP2000is employed to establish a threedimensional finite element model, the nonlinear behavior of the structure was taken intoaccount. And the modal analysis of this model is taken.2. Considering uncertainties of earthquake and structural model and based on cloudmethod, seismic fragility curves of bridge components and overall bridge structure areobtained.3. After obtaining the needed sample point using Latin Hypercube Sampling (LHS)approach and Icremental Dynamic Analysis (IDA) approach, two methods based onRBFNN are used to get fragility curves of structural components and overall structure.4. After obtaining the needed sample point using LHS approach and IDA approach,two methods based on SVM are used to get fragility curves of structural components andoverall structure.It turns out that, the fragility analysis results of the method based on Cloud Method and RBFNN are not matched well in the longitudinal direction but are matched well in thelateral direction. And on condition that the same or even higher computational accuracy isgot, the methods based on SVM can not only reduce computational work effectively, andalso work better than the methods based on RBFNN.