MS-11-163-cd banner.jpg Synthesis of Design Ground Motion for Nonlinear Analysis with Features Identified from, Records Satisfying Specified Condition Di Lin1, and Riki Honda2 1Department of Civil Engineering, University of Tokyo, Japan 2Department of International Studies, University of Tokyo, Japan ABSTRACT

For the seismic design of structures considering extremely strong ground motions, input ground motions that are influential for nonlinear response are essential. Today, large number of ground motions are available, but it is difficult to generate input motions that satisfy the specified design condition and simultaneously be significantly influential for nonlinear response. We propose a method to generate such input motions by extracting the features of the members of candidate ground motions by autoencoder (machine learning) and synthesize the time series by modification of the values of features. The method is applied the case assuming the target linear response spectra, considering the nonlinear behavior of the target structure.

Keywords: Autoencoder, seismic design, ground motions, nonlinear response, feature identification

 pdflogo.jpg MS-11-163