A Probabilistic Predictive Model for Foundation Settlement on Liquefiable Soils Improved with Ground Densification

Abstract

In this paper, we present a probabilistic predictive procedure for a foundation’s permanent average settlement on liquefiable soils improved with ground densification. The proposed procedure is based on 770 three-dimensional (3D), fully coupled, effective-stress, finite-element analyses designed through quasi-Monte Carlo sampling of key input parameters. The numerical models are themselves calibrated and validated with centrifuge model studies, and they consider realistic, nonlinear, 3D structures on shallow foundations, seismic soil–structure interaction, interlayering and layer cross interactions, ground densification properties and geometry, and ground motion characteristics. We use nonlinear regression with lasso-type regularization to estimate model coefficients. The primary predictors of a foundation’s settlement are identified as the cumulative absolute velocity of the outcropping rock motion; total thickness of the soil deposit above bedrock and cumulative thickness of the critical liquefiable layer(s); the foundation’s bearing pressure, size, and embedment depth; the structure’s total height; the achieved density and size of ground improvement; and the thickness of the remaining undensified susceptible soils within the foundation’s influence zone. In the end, the predictive model is shown to capture the trends in a limited number of centrifuge and field case histories collected from the literature. The insight from the numerical database and the first-of-its-kind predictive model aims to guide the design of liquefaction mitigation strategies that improve the performance of the soil–foundation–structure system holistically and reliably.