Integrating Finite Element Modeling and Machine Learning for Hydro-Mechanical Analysis of Swelling in Clay-Sulfate Rocks

The swelling behavior of clay-sulfate rocks poses significant challenges in geotechnical engineering due to complex hydro-mechanical and chemical interactions. This study introduces a hybrid framework combining finite element modeling (FEM) with machine learning (ML) to efficiently analyze and predict the nonlinear behavior of swelling processes. We generated synthetic datasets representing swelling phenomena at the Staufen site, Germany, using a coupled hydro-mechanical FEM simulation in OpenGeoSys. A parametric analysis was conducted to systematically vary critical parameters, including Young's modulus, permeability, maximum swelling pressure, and air entry pressure, to capture the inherent uncertainty and variability of swelling processes. A constrained CatBoost ML model was trained on the FEM outputs to predict porosity, saturation, and displacement under varying hydro-mechanical conditions. Results demonstrate strong alignment between the ML surrogate and FEM simulations, achieving high accuracy while significantly reducing computational demands. Sensitivity analysis indicated the dominance of swelling pressure and Young's modulus in influencing swelling-induced deformation, while Monte Carlo simulations quantified prediction uncertainties. This contribution discusses the potential of integrating FEM with ML for site-specific risk assessment and mitigation planning in geotechnical engineering.