Topic 4: Machine Learning in Geotechnical Engineering Topic 4: Machine Learning in Geotechnical Engineering
Challenge
- Traditional methods for predicting soil properties are time-consuming and computationally intensive.
- Modeling soil mechanics accurately requires sophisticated constitutive models, but their calibration are costly.
Contribution
- Compiled a generalized soil properties database and applied machine learning methods to predict them.
- Proposed a physics-informed neural network (PINN) model to reproduce the hydromechanical behavior of unsaturated soils.
- Developed a data-driven model for frozen soils, using Monte Carlo dropout to assess uncertainty and enhance its reliability.
Back