TITLE:
Machine Learning-Based Constitutive Models for Soil-Water Retention, Hydraulic Conductivity, and Shear Strength of Unsaturated Saprolitic Soils
AUTHORS:
Paulo Mauricio Silva Lopes, André Luis Brasil Cavalcante, Juan Manuel Girao Sotomayor, Patrícia Figuereido de Sousa, Vidal Félix Navarro Torres, Giovana Abreu de Oliveira
KEYWORDS:
Unsaturated Soils, Machine Learning, Constitutive Models, Soil-Water Retention Curve, Hydraulic and Mechanical Behavior
JOURNAL NAME:
Journal of Geoscience and Environment Protection,
Vol.13 No.6,
June
27,
2025
ABSTRACT: The mechanics of unsaturated soils is a relatively recent and evolving field of study. This paper introduces an innovative machine learning-based approach for developing constitutive models to describe the soil-water retention curve, hydraulic conductivity, and shear strength of unsaturated soils. These models were built using comprehensive soil characterization data and triaxial test results, incorporating parameters such as gravel, sand, silt, and clay content, plasticity index, porosity, and permeability. Equations were implemented using algorithms developed in the Mathematica® programming environment. The results demonstrate that the proposed models are both physically consistent and experimentally validated, exhibiting high precision and practical applicability. While this approach significantly optimizes the development of constitutive models, it does not replace the need for conventional testing, instead serving as a robust complementary tool. The proposed methodology offers an efficient and reliable solution for generalizing constitutive models across various unsaturated soil types, advancing knowledge and applications in the field.