Deep Learning for Multiscale Models
Project, University of Notre Dame, 2025
- Guide: Dr. Zhiliang Xu, Professor, ACMS Department, University of Notre Dame, Notre Dame, IN.
- About the project: This work is part of my Graduate Research Assistantship at the University of Notre Dame.
- Architected and implemented an Energetic Variational Deep Neural Network (EVNN) solver in PyTorch to model Cahn-Hilliard phase-separation dynamics.
- Ensured model stability and physical consistency by enforcing energy conservation laws directly within the neural network architecture, resulting in more robust and reliable simulations.
- Scaling this EVNN framework to model complex, coupled Cahn–Hilliard–Navier–Stokes systems to improve training stability for high-dimensional fluid dynamics.
