Reliability-Based Sensitivity Analysis of Tailings Dam Resistance Parameters

Abstract

The inherent variability of soil parameters challenges traditional deterministic stability analyses, which often fail to quantify uncertainty. Reliability analyses are therefore recommended to account for input uncertainties, although they can be computationally demanding. In this context, sensitivity analyses help identify the parameters that most strongly influence the variability of the structural response. This study conducts a sensitivity analysis of resistance parameters for a mining dam near Belo Horizonte, Minas Gerais, Brazil. Two complementary methods were employed: Sobol indices, based on the Kucherenko approach and the First-Order Reliability Method (FORM), which provides sensitivity factors at the design point. Additionally, active learning surrogate modeling techniques, in combination with Polynomial Chaos Expansion (A-PCE) and Kriging (A-Kriging), were used to reduce computational demand while maintaining accuracy. According to FORM, the effective friction angle ϕ′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {\phi{\prime} } \right)$$\end{document} of the residual soil was identified as the dominant parameter, with a sensitivity factor of 0.83, followed by the effective cohesion c′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {c{\prime} } \right)$$\end{document} of the residual soil, with a sensitivity factor of 0.10. The Kucherenko approach coupled with surrogate modeling also identified the ϕ′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi{\prime}$$\end{document} of the residual soil as the most influential parameter, with total sensitivity indices ranging from 0.43 to 0.51, while the ϕ′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi{\prime}$$\end{document} of the embankment exhibited a significant secondary contribution, with indices ranging from 0.38 to 0.45. In contrast, the filter strength parameters and the unit weight of all soil layers presented negligible sensitivity indices. Complementing the sensitivity analysis, reliability assessments were performed using both the FORM method and Subset Simulation (SS) under the surrogate modeling framework, with all methods producing similar results and demonstrating equivalence in this case study.