Sungwon Kim

I am Sungwon Kim (pronounced “Sung-won”), a Ph.D. candidate in the Graduate School of Data Science (GSDS) at KAIST, where I am advised by Prof. Chanyoung Park. I hold a B.S. degree in Civil, Environmental and Architectural Engineering from Korea University.

I am actively engaged in research with my colleagues at the Data Science and Artificial Intelligence Lab.

---

🔬 Core Research Focus

AI Surrogate Modeling for CAE and PDE solvers (Neural Operators)

My primary research focuses on developing high-fidelity AI surrogate models that work in synergy with computationally intensive 3D CAE simulations (e.g., structural mechanics, fluid dynamics, injection molding analysis) and PDE solvers (i.e., neural operators). Rather than replacing these established tools, my models augment them to accelerate engineering design cycles and lower computational costs.

Keywords: Physics AI (Engineering), 3D Simulation, Physics-Informed Neural Networks (PINNs), Neural Operators

Key Focus:

• Scalability: Building surrogate models that scale to industrial-level 3D problems with high resolution and geometric complexity.
• Usability: Streamlining surrogate models into practical engineering workflows for real-world adoption.
• Geometry-Generalizability: Developing models that generalize robustly across diverse and unseen 3D geometries.

Projects

• Physics-AI, Learning-based 3D Simulation (Collaboration with LG Electronics)
  • How can we develop a transformer-based model that operates at high scalability (industrial level)?
  • How can we create a learning-based alternative to the FEM for 3D inputs with highly complex geometries, given the initial and boundary conditions? (Point-cloud based)
  • How can we efficiently interact with opposing surfaces while maintaining computational efficiency? (Mesh based)