Radial Basis Function Neural Network-Based Method For Seismic Fragility Analysis Of Steel Frames

Building structures require seismic risk and reliability assessment which is an uncertain criterion.The risk of failure due to earthquakes is prominent in steel-framed structures.Usually,seismic fragility analysis is conducted to stay on the safe side with the probabilistic point of view.The initial phase consists of demarcating the uncertainty in the material,geometric,and earthquake parameters.The seismic fragility analysis is completed using ‘Cloud’ approach;this approach requires non-linear response history analysis.Enormous duration is required for several time-history analyses that increase the computational cost exponentially.The overall process of obtaining damage index after such analysis is rigorous.Machine learning techniques like artificial neural networks are emerging in the soft computing field.Previous researches have shown that radial basis function neural network(RBFNN)can predict the seismic damage with enough data.However,there is a lack of comparison among different neural network architectures with different configurations.Therefore,it is further necessary to figure out the best possible neural network and its architecture for faster prediction of seismic fragility curves.This dissertation conducts the following studies based on the concept of machine learning.(1)It introduces selection and uniform distribution of parameters and properties of steel frame structure for conducting simulations.Uniform Design Method(UDM)is used for uniform sampling of the multi-factored and multi-leveled experiment in this study.Also,detailed description of Finite Element(FEA)setup of steel frames is described.(2)It introduces the step by step procedure to perform cloud analysis and develop the seismic fragility curves.Park-Ang damage index is used as a structural damage indicator.Such data obtained from the nonlinear time history analysis is later used for training,testing and validation of artificial network(ANN)to predict the fragility curves.(3)It introduces the radial basis function neural network(RBFNN)method for the prediction of damage index which is used to derive seismic fragility curves.Among the 4-different architectures,the neural network with spectral displacement(Sd)as an input parameter stood as an outstanding performer.Fragility curves of both predicted and FEA results are compared with each other and the previous findings.The obtained results indicate that ANN can predict the fragility curves satisfactorily when the amount of data is less.Lastly,more data is required for higher precision in obtaining fragility curves.