Using field inversion (FI) and symbolic regression (SR) to turn high-fidelity data into explicit, closed-form models: a mechanism-separated transition model for RANS (under review, AIAA Journal) and a vortex-aware reconstruction selector for shock-capturing schemes (under review, JCP).
An FP32 GPU finite-volume RANS/LES solver built around an exact discrete adjoint whose residual derivatives are generated by CuAD — a source-to-source automatic-differentiation compiler for CUDA. Drives gradient-based shape optimization, Cf/Cp inverse design, and Spalart–Allmaras field inversion (FIML) from Python, with the adjoint linear solve in double precision through PETSc. Validated on Schubauer–Klebanoff transitional flat plate, Mach-7 axisymmetric shock/boundary-layer interaction, and the ONERA M6 transonic wing.
An open-source, Tapenade-style source-to-source AD compiler for CUDA built on Clang LibTooling. It generates exact forward (tangent) and reverse (adjoint) GPU kernels from the original CUDA source, and powers the discrete adjoint of CuDAFLOW.
Solver-agnostic structured overset (chimera) mesh generation for rotors: wall-resolved (y+ ≈ 1) blade boundary-layer meshes from a plain-text planform table, full-rotor replication, and matched Cartesian background grids. Released on GitHub with a Zenodo DOI.
Comprehensive aerodynamic characterization of various eVTOL configurations and validation of the Lattice-Boltzmann Method (LBM) for near-ground operations.
Propeller–wing interaction modeling and wing design optimization within AMADEVS, a generalized aircraft modeling and multidisciplinary design environment developed at AVDL, SNU.
Developing a high-performance M&S environment and Coaxial rotor analysis module for advanced rotorcraft sizing and optimization.