These papers reflect my work on graph neural networks for CFD, spatio-temporal forecasting, multi-fidelity Bayesian learning, UAV controller optimization, and drone acoustic analysis.
A GB-AE-GCN architecture with gradient-based pooling and Mahalanobis reconstruction for aerodynamic prediction on complex unstructured meshes. The documented workflow reports over 99% computational savings compared with CFD data generation.
A simulation-driven pipeline for tuning quadrotor PID gains across multiple objectives. The study reports Grey Wolf Optimization as the strongest method and shows a 42.7% cost reduction on an unseen mission in simulation.
A contribution focused on graph-convolutional autoencoder frameworks for aerodynamic surface field prediction on the AGARD wing. Public metadata is available, while the full publisher-hosted paper remains access-restricted.
A research contribution connecting drone acoustics and psychoacoustic annoyance prediction through neural models, extending the broader line of work around quieter and more efficient UAV operation.
An AeroNet framework that compresses pressure fields and forecasts aerodynamic dynamics with graph-based temporal layers. The cited workflow also reports over 99% computational savings relative to CFD.
A multi-fidelity Bayesian neural network that fuses aerodynamic datasets with different fidelity levels while estimating uncertainty, outperforming Co-Kriging on the reported high-fidelity test case.
Download portfolio