Development of an inverse analysis framework for extracting dynamic soil behavior and pore pressure response from downhole array measurements

Abstract

SUMMARYObservations from earthquakes over the past several decades have highlighted the importance of local site conditions on propagated ground motions. Downhole arrays are deployed to measure motions at the ground surface and within the soil profile, and also to record the pore pressure response within the soft soil profiles during excitation. The degradation of soil stiffness as excess pore pressures are generated during earthquake events has also been observed. An inverse analysis framework is developed and demonstrated to directly extract soil material behavior including pore water pressure (PWP) generation from downhole array measurements that can then be readily used in 1D nonlinear site response analysis. The self-learning simulations (SelfSim) inverse analysis framework, previously developed for total stress site response analysis, is extended to extract PWP generation behavior of the soil in addition to cyclic response during ground shaking. A Neural Network based constitutive model is introduced to represent PWP generation during cyclic loading. A new analysis scheme is introduced that can use data from co-located piezometer and accelerometer sensors. The successful performance of the proposed framework is demonstrated using four synthetic vertical array recordings.

Influence of Sampling Methods on the Accuracy of Machine Learning Predictions Used for Strain-Dependent Slope Stability

“…Supervised machine learning (ML) methods like the artificial neural network (ANN) or multi-layer perceptrons (MLP) have been successfully applied to diverse geotechnical engineering problems. Examples of the application of supervised machine learning methods in geotechnics date back more than 25 years ago and include, for example, settlement estimation due to tunneling [1][2][3], the estimation of pile bearing capacity [4][5][6][7][8], foundation settlements [4,9], slope stability analysis [10][11][12], liquefaction potential assessment [4,13] and the adjustment of soil model properties to match field or experimental observations [14][15][16][17][18]. Among others, comparison and review of different ML algorithms has been conducted in [19,20].…”

Identification of soil dynamic properties of sites subjected to bi-directional excitation

“…This work used a global optimization scheme to estimate low-strain soil properties of instrumented sites (Assimaki et al 2006;Assimaki and Steidl 2007;Assimaki et al 2008). Tsai and Hashash (2008;2009), on the other hand, and more recently Groholski et al (2013;, implemented an inverse analysis framework, referred to as self-learning simulations (SelfSim), that uses downhole array data during the shaking of a site to develop a neural network-based material constitutive model. Most recently, Mercado et al (2015) incorporated the methodology for estimation of shear stresses and strains, proposed by Zeghal, Elgamal and coworkers, to introduce an alternative nonlinear technique to characterize the shear stress-strain response.…”

Characterization of the contractive and pore pressure behavior of saturated sand deposits under seismic loading

“…Yang and Elgamal [19] applied an optimization analysis for the calibration of a multi-surface plasticity soil model that handles the coupling effects of soil behavior and pore pressure buildup. More recently, Groholski and Hashash [20], and Groholski et al [6] extended the SelfSim framework to effective-stress considerations in order to extract both soil behavior and pore pressure response from recorded motions and pore pressures during seismic events. This paper presents a technique to identify the coupled shear-volume response of sand deposits that can lead to significant generation of pore water pressures.…”