Electronic Theses and Dissertations
GAINING SCIENTIFIC AND ENGINEERING INSIGHT INTO GROUND MOTION SIMULATION THROUGH MACHINE LEARNING AND APPROXIMATE MODELING APPROACHES
Date of Award
Doctor of Philosophy
This dissertation presents a series of methods for gaining scientific and engineering insight into earthquake ground motion simulation in three areas: synthetic validation, attenuation modeling, and nonlinear effects estimation. First, I present guidelines to reduce the number of metrics used to evaluate the goodness-of-fit (GOF) between ground motion synthetics and recorded data in an application independent framework. Validation of ground motion simulations is mostly done using metrics that are user- or application-biased. Comparisons between synthetics from regional scale ground motion simulations and recorded data from past earthquakes provide opportunities to approach the problems using data-driven methods. I used a combination of semi-supervised and supervised learning methods to prioritize GOF metrics based on a large dataset and was able to identify the response spectra- and energy integral-based metrics as the most dominant ones for estimating the accuracy of simulations. Second, in two related studies, I present an application of customized solutions used to characterize attenuation (quality factor Q) with respect to shear wave velocity (Vs) for individual stations within a simulation. I used an artificial neural network as a supervised learning method to develop pseudo-simulators to be used in an optimization process to estimate the dominant Vs range for each station, and thus estimate Q. Using parameters such as peak ground acceleration, response spectra, the area under the velocity signal's envelope and the peak ground velocity, I show it is possible to improve the optimization process to locate the most accurate Q parameters. Last, I present an approximate model to estimate nonlinear soil effects in ground motion simulations by implementing an approach inspired in the equivalent linear method. This implementation is done for three-dimensional simulations, from source to site, without any pre- or post-processing of data. Fully nonlinear ground motion simulation methods need comprehensive input data and are computationally challenging. The approach implemented can be used to estimate first-order nonlinear soil effects (e.g., deamplificaiton and resonant frequency shift) effectively. I calibrate the approach using idealized models.
Dissertation or thesis originally submitted to ProQuest
Khoshnevis, Naeem, "GAINING SCIENTIFIC AND ENGINEERING INSIGHT INTO GROUND MOTION SIMULATION THROUGH MACHINE LEARNING AND APPROXIMATE MODELING APPROACHES" (2018). Electronic Theses and Dissertations. 2919.
Available for download on Saturday, May 25, 2024
Data is provided by the student.