A MACHINE LEARNING APPROACH TO NONLINEAR RESPONSE ANALYSIS OF STRUCTURES

摘要

Nonlinear response history analysis (NRHA) is the most accurate seismic performance assessment method but it can be computationally intensive when it is applied to multi degree of freedom (DOF) structural systems. Machine learning techniques have gained increasing interest in engineering fields and can consist a powerful tool for reliable predictions with their ability to quickly and accurately identify trends or patterns through experimental or artificially generated data. In this study, we propose a robust machine learning pipeline to estimate the nonlinear response analysis of multi-DOF systems in terms of their maximum displacement/ductility aiming to eliminate the computational cost of the NRHA analysis. A pulse extraction process is used to quantify the wavelet parameters of a ground motion records which, along with the material parameters of the structural system consist the training data set. It is shown that adequate predictions were obtained through the validation of various benchmark structures which can act as a reference tool for an engineer in practice.