Learning from Real-World Data to Inform High-Fidelity Modeling: Bayesian Model Updating for Identification of Soil-Structure-Interaction Dynamics Considering Material Nonlinearity
The dynamic response of a building structure to an earthquake excitation is the result of a complex interaction between the structural system and the underlying soil and surrounding geology. Since modeling the physics of a coupled soil-structure system in detail is an intricate undertaking, the state-of-practice has adopted simplified modeling and analysis procedures, such as the substructure approach. In this approach, the soil flexibility and energy dissipation are modeled using distributed spring and dashpot elements to which the foundation input motions (FIMs) are applied as uniform base excitations.
The stiffness and viscosity of these elements are derived using simplified analytical methods, which are nonetheless based on idealized and restrictive assumptions – e.g., linear-elastic response behavior of soil and structure, uniform soil half space, etc. These simplifying assumptions and the empirical nature of the mechanical analogs (e.g., soil springs and dashpots) could potentially lead to unquantified errors in predicting the seismic responses of real-world building structures, even though the simplified models have demonstrated acceptable accuracy in idealized case studies. Furthermore, integrating the substructure method with the Rayleigh damping model – which itself has a highly empirical nature –introduces another source of uncertainty in the seismic analysis of buildings. Clearly, there is an inconsistency between the relatively high sophistication level of mechanics-based techniques available for structural system modeling and the crude simplicity of the underlying assumptions that guide the damping and soil-structure interaction (SSI) models. This is an important shortcoming in the current state-of-practice that can significantly affect the seismic response simulation and risk assessment of critical buildings. This project aims at addressing this important shortcoming using a new approach. The proposed approach is based on distilling information from real-life data to guide efforts on damping and SSI model characterization.