Dynamic site response analysis in the face of uncertainty–an approach based on response surface method

Although advanced modeling techniques such as the finite element method (FEM) have been used successfully in dynamic site response analysis, the high computational expense has hindered the incorporation of input parameter uncertainty in such analysis. Thus, when the variation of the peak ground acceleration (PGA) at the ground surface, which is the outcome of the site response analysis, has to be evaluated in the face of uncertainty, a surrogate model such as Response Surface Method (RSM) model is often used in lieu of the FEM model. In this paper, the RSM surrogate model was implemented in the context of seismic site response analysis to evaluate the variation (or uncertainty) of the PGA due to the propagation of input parameter uncertainty. The engineering implication of the variation of the surface PGA is significant as this knowledge is required for such tasks as reliability analysis of soil liquefaction and probabilistic seismic risk analysis. To derive the RSM model specifically for a site response analysis, a parametric study of the dynamic site responses using FEM code ABAQUS with the modified Davidenkov soil constitutive model implemented as a user-defined material subroutine is first conducted. The input parameters, including the soil profile, soil properties, and input ground motion, in a typical dynamic site response analysis are then characterized and screened for their suitability to be included in the RSM model. For a given site response problem in the face of uncertainty, representative “samples” are taken for site response analysis using ABAQUS, and the analyses are carried out and finally the response surface is constructed using the outcomes of the ABAQUS numerical experimentations. The accuracy of the developed RSM model is then validated using the results of ABAQUS on cases not used in the development of the RSM model. Once the RSM model is deemed satisfactory, the reliability-based formulation for evaluating the mean and standard deviation of the surface PGA is established. Example is provided to illustrate the RSM approach for dynamic site response analysis in the face of input parameter uncertainty.