RESPONSE SPECTRUM BASED GENERATION OF SPATIALLY VARYING EARTHQUAKE USING ARTIFICIAL NEURAL NETWORKS

This paper presents a procedure based on artificial neural networks that is capable of generating spatially varying accelerograms from a single response spectrum. Non-uniform ground motions may have influence on the seismic response of structures that have spatially extended foundations. Artificial spatially varying earthquakes will be required for more nonlinear analyses to consider this effect. Learning capabilities of two types of Artificial Neural Networks (ANNs) are used to develop a method for generating spatially varying accelerograms from response spectra and distance parameter. Recorded accelerograms from SMART-1 array in Loting, Taiwan were used to train neural network. Generated artificial spatially varying earthquakes can be used for time history analysis of large structures such as bridges and dams.