Files in this item

FilesDescriptionFormat

application/pdf

application/pdfILHAN-DISSERTATION-2020.pdf (89MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Conventional and deep learning-based site amplification models for central and eastern North America
Author(s):Ilhan, Okan
Director of Research:Hashash, Youssef M. A.
Doctoral Committee Chair(s):Hashash, Youssef M. A.
Doctoral Committee Member(s):Olson, Scott M.; Elbanna, Ahmed E.; Stewart, Jonathan P.; Rathje, Ellen M.; Nikolaou, Sissy
Department / Program:Civil & Environmental Eng
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Site Amplification, Deep learning, Site effects, Ergodic model
Abstract:Central and Eastern North America (CENA) is known as a stable continental region that lacks region-specific site amplification functions due to a dearth of ground motion recordings for empirical constraints. Recently developed site amplification functions using a parametric study of large-scale one-dimensional (1D) site response analyses represent significant improvement in terms of seismic hazard applications. However, the limitations of these simulation-based studies persist due to (1) constraints on the variability in site profiles produced by parametric study and (2) the large error of estimations by conventional amplification functions caused by forcing a priori functional forms to fit the data. This research aims to develop robust response spectrum (RS) and Fourier amplitude spectrum (FAS) amplification models using an enhanced and enlarged amplification database. The first segment of this study is to perform over 3.6 million 1D linear, equivalent-linear and nonlinear site response analyses via high performance computing (HPC) resources. The simulations are conducted using prior parametric study design with improvements to remove the constraint on the variability of simulated site conditions at CENA along with the bias in the generic profiles. This study provides a brief description of each element of the parametric study tree and a detailed explanation of updates relative to prior work. The second part of this research is the development of RS and FAS site amplification models through a deep learning-based approach via Artificial Neural Network (ANN) and the application of prior conventional functions with enhancements to better capture the simulated amplification data. This study is the pioneer in the implementation of ANN for the modeling of site amplification and shows that the significant decrease in error of estimations by ANN-based models is obtained relative to conventional relationships along with more accurate representation of the features observed in the amplification dataset. A new conventional relationship, based only on the site natural period, which produces noticeably lower errors relative to those using time averaged shear wave velocity in the top 30 m of a site (VS30) is proposed for linear RS amplification, and all the linear RS functions are applied to FAS amplification to suggest an alternative set of models. The standard deviation for ANN-based and conventional models are proposed, and the improvements of ANN over conventional functions in capturing site-specific response are discussed.
Issue Date:2020-10-12
Type:Thesis
URI:http://hdl.handle.net/2142/109480
Rights Information:Copyright 2020 Okan Ilhan
Date Available in IDEALS:2021-03-05
Date Deposited:2020-12


This item appears in the following Collection(s)

Item Statistics