Enhancing seismic site response analyses: tuning soil properties via genetic algorithms and Bayesian model updating from downhole array data

Abstract

Abstract This study employs deterministic genetic and probabilistic Bayesian algorithms to enhance seismic site response simulations through refined characterization of soil properties using downhole array data. We developed a three-dimensional finite element model using OpenSeesPy software, which incorporates isotropic elastic soil material model and Lysmer-Kuhlemeyer dashpots to simulate radiation damping. This model adjusts shear wave velocities across different soil layers, aiming to minimize the discrepancies between observed and predicted acceleration response spectra at various depths. By utilizing data from the 2011 Kütahya earthquake (5.8 Mw) recorded in Istanbul’s Zeytinburnu district, the framework was calibrated and validated. This data encompasses measurements from bedrock, two mid-layers and the surface. Initial shear wave velocities for these layers were established based on average values derived from PS Suspension Logging tests. Updated model parameters are then calculated as multiples of these initial estimates, with the modification bounds defined by the standard deviations of the initial parameters. Finally, the so-updated parameters are used in the model to validate the response against a the Ege Denizli (Aegean Sea) earthquake (6.2 Mw). The practical application of this model demonstrates its capability not only to align closely with empirical seismic data, thus enhancing the accuracy of predictions, but also to effectively quantify uncertainties associated with seismic site response, showcasing the robustness of the combined deterministic and probabilistic approaches in real-world settings..