Physics-informed semi-empirical probabilistic models for predicting building settlement and tilt on liquefiable ground

S. Dashti

This presentation introduces predictive models for the seismic settlement and tilt of shallow-founded structures on liquefiable ground based on an integrated observational, experimental, numerical, and statistical approach. Effective liquefaction mitigation requires an improved understanding of the consequences of liquefaction on structures. The state of practice typically involves estimating building settlement using empirical procedures for free-field conditions, which have been shown to be unreliable and inappropriate through previous case histories and physical model studies. To address this problem, first, a series of centrifuge experiments were performed to evaluate the dominant mechanisms of deformation near shallow-founded structures. Second, experimental results were used to evaluate the predictive capabilities of 3D, fully-coupled, nonlinear, dynamic finite element analyses of soil-structure systems in OpenSees. Third, a numerical parametric study (exceeding 63,000 simulations) was used to identify the most optimum Intensity Measures for permanent building settlement and tilt as well as the functional form of predictive models. And finally, a case history database helped validate and refine the models, accounting for field complexities not captured numerically or experimentally. This integrative approach yielded a set of procedures that are the first to consider variations in soil  layering and geometry, foundation and structure properties (in 3D), soil-structure interaction, contribution of all mechanisms of deformation, and total inherent model uncertainties—all of which are necessary to realize the benefits of  performance-based seismic design in evaluating and mitigating the liquefaction hazard. These probabilistic models are first introduced for individual buildings, followed by their extension to regional analysis and mapping.