projects
Main research directions and ongoing work in machine-learning-enabled CFD and industrial simulation.
Overview
My work sits at the intersection of fluid dynamics, machine learning, and scientific computing. I focus on building data-enabled methods that make industrial simulation faster, more reliable, and more useful in engineering design.
Across recent projects, the common goal is to reduce turnaround time for aerodynamic analysis and turbulence modeling while preserving the physical fidelity required in real engineering workflows.
Real-Time Aerodynamic Evaluation
I develop machine-learning-based operators and geometry-aware models for rapid aerodynamic prediction on complex three-dimensional configurations. The objective is to move from expensive repeated CFD loops to near real-time design feedback.
This direction is especially relevant for industrial applications where large design spaces must be explored under tight iteration cycles.
- Geometry-to-flow learning for arbitrary 3D vehicle configurations.
- Fast surrogate models for drag and aerodynamic performance estimation.
- Bridging high-fidelity CFD data and practical engineering decision-making.
Data-Enabled Turbulence Modeling
A major part of my research studies how data and machine learning can improve turbulence closures while preserving physical consistency. This includes screening candidate models, recalibrating classical closures, and designing workflows that remain interpretable and robust.
The emphasis is not only predictive accuracy, but also deployability in realistic Reynolds-averaged and large-eddy simulation pipelines.
- Data-enabled recalibration of RANS turbulence models.
- A priori screening and constrained model development.
- Large-language-model-assisted turbulence model exploration.
Reduced-Complexity Solvers and Industrial CFD Tools
I am also interested in reducing the computational cost of iterative solvers and simulation workflows used in production environments. The goal is to improve turnaround time without sacrificing numerical reliability.
This direction connects numerical methods, software tooling, and engineering deployment, especially for industrial users who need stable and efficient simulation systems rather than isolated algorithmic prototypes.
- Data-enabled reduction of solver time complexity.
- Practical software pipelines for industrial aerodynamic simulation.
- Integration of AI components into CAE and CFD workflows.