Mechanical Engineering
NOTICE OF 2nd PhD SEMINAR PRESENTATION
CANDIDATE: John Agyapong Boamah
DEGREE SOUGHT: PhD
DATE: 10/2/2026
TIME: 11:30am
LOCATION: Room 1101 CEI
TITLE: Transfer learning for reconstruction of missing regions in a near-wall cylinder wake.
Abstract
Reconstructing missing regions in velocity fields is a key challenge in experimental fluid mechanics, particularly for complex turbulent flows with limited spatial resolution. This study presents a transfer learning framework that combines large-eddy simulation (LES) and particle image velocimetry (PIV) data to reconstruct incomplete wake velocity fields. A linear extended proper orthogonal decomposition (EPOD) model is compared with a nonlinear U-Net under controlled masking conditions. The U-Net is pretrained using time-resolved LES data and fine-tuned with non-time-resolved PIV measurements obtained under the same flow conditions. Results demonstrate that the nonlinear model provides improved reconstruction of spatial flow structures and velocity correlations compared with the linear baseline. Transfer learning provides modest improvements in reconstruction accuracy and agreement with Reynolds shear stress, although these gains are less consistent for higher-order flow statistics. The results show the potential of CFD-informed pretraining for improving velocity-field reconstruction while emphasizing the challenges of recovering multi-scale turbulent dynamics from experimental measurements.
All Graduate Students are invited to attend