Global Seismic Reliability Analysis Of Reinforced Concrete Frames Based On ANN And Number-theoretic Point-selection

During the entire life of engineering structures,there are many kinds of loadings,whose duration and intensity are stochastic.Meanwhile,there are uncertainties in structures.These uncertainties lead to the uncertainties of structural responses.Therefore,it is necessary to analyze the failure probability of structures using the reliability method.In this thesis,the artificial neural network and number theory are combined to construct the surrogate model of structural global limit state function,which is used to analyze the reliability.The main research contents are as follows:(1)Based on the reliability analysis steps of ANN and the number theoretic point-selection,the programs are written by MATLAB.Chosing some explicit expressions and a portal frame as examples,the reliability indices are calculated by the Monte Carlo simulation method,the artificial neural network with Monte Carlo simulation and the artificial neural network with number theoretic point-selection.Then,the results are compared to check the accuracy and efficiency of the NT-ANN.(2)Chosing the three-storey RC frame as the research object,the limit state functions of the load carrying capacity and the deformation limit state are constructed with the structural global demand and capacity.Then the static reliability indices of these two limite states are calculated by MCS,MCS-ANN and NT-ANN combined with the nonlinear static method and the capacity spectrum method.(3)On the basis of the static reliability analysis of the deformation limit state,the certainty ground motion and artificial ground motion with temporal and spectral non-stationary are chosen as the input loading.Then the dynamic reliability indices of the deformation limit state are calculated by NT-ANN and MCS.(4)Using the fatigue failure criterion,the global seismic damage limit state function is construct.Then the dynamic reliability indices are calculated by MCS and NT-ANN with the certainty and artificial ground motion input.The results are contrasted to vertify the NT-ANN method.Through the study of this thesis,it can be concluded that when the sample point generated by the suitable creating vector,the surrogate model constructed by NT-ANN can replace the structure model with the high precision.Comparing with the MCS and MCS-ANN,the calculation cost of NT-ANN is much less with the same accuracy.