Preliminary Design of Seismic Isolation Systems Using Artificial Neural Networks

Abstract

This works attempts to implement artificial neural networks (ANN) for modeling Seismic-Isolation (SI) systems consisting of Natural Rubber Bearings and Viscous Fluid Dampers subject to Near-Field (NF) earthquake ground motion. Fourteen NF earthquake records representing two seismic hazard levels are used. The commercial analysis program SAP2000 was used to perform the Time-History Analysis (THA) of the MDOF system (stick model representing a realistic five-story base-isolated building) subject to all 14 records. Different combinations of damping coefficients (c) and damping exponents (α) are investigated under the 14 earthquake records to develop the database of feasible combinations for the SI system. The total number of considered THA combinations is 350 and were used for training and testing the neural network. Mathematical models for the key response parameters are established via ANN. The input patterns used in the network included the damping coefficients (c), damping exponents (α), ground excitation (peak ground acceleration, PGA and Arias Intensity, Ia). The network was programmed to process this information and produce the key response parameters that represent the behavior of SI system such as the Total Maximum Displacement (DTM), the Peak Damper Force (PDF) and the Top Story Acceleration Ratio (TSAR) of the isolated structure compared to the fixed-base structure. The ANN models produced acceptable results with significantly less computation. The results of this study show that ANN models can be a powerful tool to be included in the design process of Seismic-Isolation (SI) systems, especially at the preliminary stages.