XGBoost based surrogate technique for system reliability analysis of foundation over cavity aided with bootstrapping

This paper introduces a comprehensive framework for conducting time-dependent system reliability analysis of foundation situated above cavity. It leverages advanced techniques, including Extreme Gradient Boost (XGBoost) regression and Monte Carlo Simulation (MCS), to assess the stability and performance of these foundations while considering uncertainties in various parameters. The core of this study involves the development of surrogate models using XGBoost regression to predict critical foundation response variables, viz., bearing capacity (BC) and settlement. These models incorporate variability through the use of the bootstrapping technique. The XGBoost regression models demonstrate superior predictive accuracy, boasting R² values of 0.97 for BC and 0.86 for settlement, and thus effectively capturing complex data relationships. To deal with non-normally distributed data, the study employs a robust log transformation approach with bootstrapping, ensuring the generation of reliable results suitable for analysis. The foundation's performance is rigorously assessed against BC and settlement criteria, both individually and as a system. A notable finding is the 29.62% increase in the probability of failure while transitioning from BC to settlement criteria, and an amplified change of 55.74% was observed for system reliability. This underscores the paramount importance of system reliability assessment. In addition, a time-dependent reliability sensitivity analysis is performed by introducing controlled temporal variations in BC and settlement, revealing a progressive reduction in reliability indices with time.