이창훈 연구실

Unsupervised learning model based on CycleGAN for super-resolution reconstruction of turbulence. Reconstructs high-resolution flow field from LES data with universal applicability across Reynolds numbers.

Deep learning approach using conditional GANs (cGANs) for predicting 2D decaying homogeneous isotropic turbulence. Trained on DNS datasets to capture input-output correlations with satisfactory accuracy.

Indoor flow and contaminant dispersion prediction using LES coupled with immersed boundary method. Analyzes contaminant transport from infiltration or leakage flow with GPU acceleration for real-time prediction.

Numerical simulations of flow from 20 kT explosion predicting mushroom cloud rise and diffusion. Uses anelastic approximation and density weighted variables with potential temperature for atmospheric stratification.

Study of settling ellipsoidal particles in turbulent flow. Proposes new formulation for spheroidal particle evolution equations and examines particle clustering behavior and orientation correlation with Lagrangian stretching.

Investigation of Rayleigh-Bénard convection with particles or bubbles using DNS. Studies interaction between particles/bubbles and background turbulence, revealing formation of large-scale polygonal cells.

DNS study of millimetric spheroidal bubble path instability in quiescent fluid and isotropic turbulence. Uses immersed boundary method to investigate zigzag pattern behavior and turbulence effects on rising bubbles.

DNS of thermally stratified turbulent channel flow with finite-size particles using immersed boundary method. Shows particles modify turbulence structures and suppress vertical heat transport through gravity wave effects.