Microstructure-informed constitutive modeling of granular media under multidirectional loading: From particle-scale...

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Artificial Intelligence

Background:

• Simulating granular material response under realistic loads is crucial for geotechnical infrastructure reliability.
• Natural soils have complex, non-uniform particle arrangements and face multidirectional loading (wind, seismic activity).
• Existing nonlinear constitutive models are limited by site-specific data and difficult generalization.

Purpose of the Study:

• To develop a deep-learning model for predicting granular media response under multidirectional loading.
• To overcome limitations of traditional constitutive models and generalize soil behavior predictions.
• To provide a computationally efficient method for analyzing complex soil mechanics problems.

Main Methods:

• Utilizing discrete element method (DEM) simulations to capture particle-scale behavior.
• Developing and training deep-learning models on DEM-generated data.
• Applying the models to predict granular material response under various loading paths.

Main Results:

• The deep-learning model accurately replicates complex, nonlinear relationships in granular material behavior.
• Demonstrated the model's capability to predict responses under challenging multidirectional loading scenarios.
• Achieved computational efficiency compared to traditional simulation methods.

Conclusions:

• Deep learning offers a powerful tool to enhance the prediction of granular material behavior.
• This approach has direct applications in optimizing geotechnical designs, such as wind turbine foundations.
• The developed model provides a pathway to more reliable and generalizable geotechnical engineering solutions.