Answers are grounded in this paper and cite their sources.
Understanding the behavior of unsaturated soils under thermo-hydro-mechanical (THM) loading is crucial for addressing challenges related to climate change, geohazard mitigation, and the design of infrastructure. Conventional constitutive models often fall short of capturing the complex interactions among THM processes, as they involve several assumptions and commonly require several input parameters that are not readily available or measurable. This paper presents a novel physics-informed data-driven approach utilizing a hybrid variational autoencoder (VAE) and generative adversarial network (GAN) model to predict the unified THM response of unsaturated soils. The hybrid VAE-GAN model incorporates physical constraints, including energy conservation and mass balance, to enhance accuracy and reliability. Experimental data from laboratory triaxial tests under varying THM conditions were used for training, validating, and testing the model. Key features were meticulously selected and preprocessed, including temperature, moisture content, stress state, strain rate, and soil properties. Results demonstrate that the hybrid VAE-GAN model surpasses traditional models in predicting soil behavior. For the cases examined, the root mean square error of the VAE-GAN model is 1/10th of that of the traditional alternative model for total volumetric strain and 0.85 for ultimate shear strength. The model was employed in a comprehensive sensitivity analysis to assess the effect of initial moisture content and porosity on the results. This study demonstrates the great yet unexplored potential of physics-informed data-driven models for modeling THM problems in unsaturated soils, offering a powerful tool for future research and practical applications.
No takes yet. Share an insight, caveat, or question.
Ajdari et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: