Event Date
Training-Free Design Generation through Morphing and Its Applications in Fluid Dynamics
Abstract
This seminar presents a data-driven yet training-free design methodology, Design-by-Morphing (DbM), which generates continuous and constraint-free search spaces for topology optimization, and explores its applications in fluid dynamic systems. By morphing between historic baseline designs, DbM avoids geometric constraints and minimizes the designer biases often imposed by traditional parameterization techniques. The benchmark application of DbM to two-dimensional airfoil optimization will be a main focus of the talk, utilizing the comprehensive UIUC airfoil database as a representative global design space. We demonstrate that DbM can reconstruct over 1,600 airfoil shapes from the database with high accuracy, enabling a radical and expansive search space. Furthermore, a recent advancement in DbM systematically reduces design-space dimensionality. By identifying an optimal baseline set of just 12 airfoil shapes through a strategy motivated by feature selection, we achieve reconstruction accuracy comparable to larger baseline sets while significantly accelerating convergence in multi-objective aerodynamic optimization. These results highlight DbM's potential as a cost-effective solution for complex engineering design problems.
Bio
Sangjoon "Joon" Lee is currently a postdoctoral scholar at Stanford University. His expertise is grounded in fluid mechanics and computational science, covering areas such as computational fluid dynamics and heat transfer, turbulence physics, vortex instabilities, and data-driven aerodynamic design optimization. Following his Bachelor's degrees in Engineering and Business from Seoul National University (2018), he earned his Ph.D. and M.S. in Mechanical Engineering with a designated emphasis in Computational and Data Science and Engineering from the University of California, Berkeley (2024)