Comparison of machine learning models in predicting dam stability

Abstract

Granular filters are integral to the safety and operational performance of soil embankment dams, as they effectively mitigate internal erosion and piping phenomena. These filters function as protective barriers designed to retain base soil particles while simultaneously facilitating the safe passage of the seepage water. An effective filter design will have pore sizes that are sufficiently small to prevent the migration of underlying soil particles, while allowing seepage to continuously dissipate. Failing to provide optimum filter and base particle combination caused nearly 50% of dam failures occur due to internal erosion Foster et al (2000). Concurrently, the filter material must maintain adequate coarse particles to promote free drainage and prevent excessive pore water pressure, making a clear understanding of the interactions between base soil and filter materials essential for effective filter systems in embankment dam construction. Although numerous relationships have been proposed regarding particle size distributions, the complexity of combining filter and base soil gradations has prevented the development of a perfect criterion. To better address these complexities, machine learning provides a powerful approach, enabling the analysis of nonlinear and interdependent soil property patterns using experimental datasets and helping identify the most accurate models for predicting soil behavior and stability.