ON AN ACCURATE ESTIMATION OF THE SEISMIC PERFORMANCE OF RC BUILDINGS WITH SSI EFFECTS USING GBRT

Soil-structure interaction (SSI) can be described as the joint response of the changes in the foundation-level ground motions and the structural behavior modification due to a reduced support capacity of the foundation medium. The latter two components are considered the soil-structure system (SSS). This system configuration contrasts with the traditional fixed-base support condition, widely used in common simplified practice, showing considerable differences in design and performance outcomes compared to those of its SSS counterpart. In the last decades, efforts have been made to reach a thorough understanding of the SSI phenomenon aided by the fast improvement in computing capabilities and the growth and application of new techniques. In this regard, it has gradually become more common to find specialized chapters in design standards worldwide dedicated to giving guidelines for the seismic design and assessment of buildings with interaction effects considerations. However, the recommendations for modeling and evaluating the SSS included in such standards still rely on the outcomes of oversimplified linear equivalent models of the structure and supporting soil because of the complex nature of a more rigorous analysis of the soil-structure interaction phenomenon using more elaborated models. To overcome this issue, this study focuses on looking for a correlation model that allows professionals to estimate the more realistic seismic performance of buildings that consider the SSI effects while analyzing through a Nonlinear Dynamic Procedure (NDP) based on the obtained performance from buildings linearly modeled neglecting such interaction effects while employing the Modal Response Spectrum Analysis (MRSA). An extensive building database was generated with varying numbers of stories, in-plan aspect ratios, and soil characteristics reflecting medium and soft soil conditions. These buildings were designed and assessed using OpenSeesPy, and estimation models were generated using the Gradient Boosting Regression Trees (GBRT) machine-learning technique. Results showed that a coefficient of determination (R²) up to 0.87 could be obtained from the GBRT model to estimate the realistic performance in terms of that determined from the MRSA, and that fundamental structural period T_n, the slenderness ratio, λ_str, and the dimensionless frequency, a_o, are key SSS parameters to assemble the performance estimation model.