An improved BUS approach for Bayesian inverse analysis of soil parameters incorporating extensive field data

Abstract

This study addresses the complexities encountered when integrating site-specific field data into the Bayesian inverse analysis of soil parameters in geotechnical structures. Traditional Bayesian methods, such as Markov Chain Monte Carlo (MCMC) and Bayesian Updating with Structural reliability (BUS), often struggle with the high dimensionality of geotechnical data and the random variables involved. In this paper, we propose an Improved BUS (IBUS) methodology, which leverages parallel system reliability analysis to overcome these challenges. By introducing auxiliary uniform random variables, we transform the structural reliability problem into one of parallel system reliability with multiple limit-state functions, effectively reducing the incidence of rare event complexities in component limit-state functions. Two slope examples are used to demonstrate the efficacy of the IBUS method. Our results reveal that IBUS is superior in incorporating sparse field data and remains robust when processing extensive datasets that would typically impede MCMC and BUS due to negligible likelihood function values. Additionally, the IBUS method offers a framework for examining the impact of data quantity and borehole numbers on slope reliability analysis, offering insights that can optimize borehole placement and layout. The findings confirm the IBUS method as a potent solution for Bayesian inverse problems with spatially variable soil parameters, affirming its usefulness even in scenarios involving extensive field data.