Abstract
An increasing number of downhole arrays are deployed to measure motions at the ground surface and within the soil profile. Measurements from these arrays provide an opportunity to improve site response models and to better understand underlying dynamic soil behavior. Parametric inverse analysis approaches have been used to identify constitutive model parameters to achieve a better match with field observations. However, they are limited by the selected material model. Nonparametric inverse analysis approaches identify averaged soil behavior between measurement locations. A novel inverse analysis framework, self-learning simulations (SelfSim), is employed to reproduce the measured downhole array response while extracting the underlying soil behavior of individual soil layers unconstrained by prior assumptions of soil behavior. SelfSim is successfully applied to recordings from Lotung and La Cienega. The extracted soil behavior from few events can be used to reliably predict the measured response for other events. The field extracted soil behavior shows dependencies of shear modulus and damping on cyclic shear strain level, number of loading cycles, and strain rate that are similar qualitatively to those reported from laboratory studies but differ quantitatively.
Original language | English (US) |
---|---|
Pages (from-to) | 745-757 |
Number of pages | 13 |
Journal | Journal of Geotechnical and Geoenvironmental Engineering |
Volume | 135 |
Issue number | 6 |
DOIs | |
State | Published - 2009 |
Keywords
- Dynamic properties
- Neural netwoks
- Shear modulus
- Shear strain
ASJC Scopus subject areas
- General Environmental Science
- Geotechnical Engineering and Engineering Geology