Data-Driven Evaluation and Analysis of Seismic Isolation System: Parameter Sensitivity

Abstract

Lead Core Rubber Bearings (LCRBs) are critical for seismic energy dissipation, however, their performance under pulse-like and long-duration earthquakes, common in subduction zones, requires further investigation. This study evaluates the sensitivity of LCRB systems to key parameters, isolator period, damping ratio, superstructure mass, design displacement, and yielding displacement, under challenging seismic conditions. A nonlinear MATLAB model simulates a single-degree-of-freedom base-isolated structure subjected to 126 pulse-like ground motions. Probabilistic variables, defined via normal distributions, are incorporated into a Monte Carlo framework, enabling 37,800 time-history analyses to quantify uncertainties in displacement, acceleration, and force responses. The results highlight the dominant influence of ground motion parameters (peak ground displacement, acceleration) on bearing displacement (B.Dis) and acceleration (B.Acc). Machine learning models (Random Forest, XGBoost) outperform linear methods, achieving R2 values up to 0.99 for B.Dis and B.Acc predictions (R2 = 0.9976 for B.Dis, 0.9908 for B.Acc) and low errors (MSE = 1.26, MAE = 0.57 for B.Dis), emphasizing their utility in capturing nonlinear interactions. Additionally, the study underscores the necessity of integrating ground motion characteristics with an isolator design to optimize seismic resilience, identifying the influential roles of PGV/PGD and PGA/PGV ratios on isolator behavior. Overall, the developed computational framework not only optimizes seismic isolation design but also facilitates practical decision-making through a user-friendly, GUI-driven data analysis tool.