Research

AI-enhanced simultion

Numerical simulation is a fundamental tool in geomechanics. High‑fidelity numerical methods (e.g., FEM, MPM) are computationally expensive, which limits their use in large‑scale parametric studies, optimization, and uncertainty quantification. Simplified or empirical models offer faster results but often sacrifice accuracy and generality. We leverage AI to develop physics simulators that are fast yet remain accurate and generalizable, enabling more sophisticated and advanced geomechanical analyses.

Related topics:

• AI-enhanced simulation with graph neural networks
• Neural operators and geotechnical analysis

physics-informed-machine-learning   graph-neural-networks   neural-operators   material-point-methods   finite-element-methods

Inverse modeling and uncertainty quantification

AI-enhanced simulators make computationally demanding geotechnical analyses, such as inverse modeling and reliability analysis, more practical. By combining physics-aware learned simulators with differentiable programming, we develop efficient frameworks for system identification, design optimization, model discovery, and uncertainty quantification. This enables large-scale optimization and probabilistic analyses that are difficult to achieve with conventional numerical methods. Ultimately, this aims to support more reliable engineering analysis and decision-making.

Related topics

• Differentiable graph neural network simulators for granular flows
• Accelerating uncertainty quantifiations in geohazard analysis

differentiable-programming   inverse-modeling   uncertainty-quantification   design-optimization   model-discovery

Computer vision for hazard detection

We explore computer vision techniques for automated hazard detection and infrastructure inspection. By leveraging advanced deep learning models, we aim to reduce labor-intensive manual inspections and enable faster, more consistent assessments of infrastructure conditions. Through collaborations with researchers and practitioners, we develop AI-assisted solutions for infrastructure health monitoring and natural hazard detection.

Related topics:

• Computer vision for infrastructure inspection and hazard detection

hazard-detection   computer-vision   object-detection   instance-segmentation

Agentic AI

Engineering software is moving beyond passive analysis tools toward intelligent agents that can reason, plan, and execute engineering workflows. We develop AI-powered agentic systems for geotechnical engineering by integrating large language models with numerical simulation, optimization, and scientific computing. Our vision is to build domain-aware AI agents that automate complex engineering workflows and enable more efficient, objective, and reproducible geotechnical analysis and design.

Related topics:

• An agentic-AI calibrate–diagnose–suggest workflow for soil constitutive models

agentic-ai   autonomous-systems   llm