Physics-informed data-driven aerodynamic modeling of distributed propulsion wing with extremely tiny datasets

Based on the research background of high-performance distributed electric propulsion (DEP) fixed-wing unmanned aerial vehicles (UAVs), a data-driven aerodynamic modeling method was developed to address the design challenges of propulsion/aerodynamic coupling in the distributed propulsion wing (DPW) configuration which fully integrates the distributed propulsors and wing surfaces. To overcome the issues of limited data and insufficient understanding of physical mechanisms in practical engineering, physical prior knowledge was introduced, and a physics-informed grey-box aerodynamic model, named Focus-VI, was established based on extremely tiny datasets utilizing variational inference and attention mechanisms. Compared with the fully black-box multi-layer perceptron (MLP), Focus-VI demonstrated advantages in higher accuracy, stronger generalization, and certain interpretability. Therefore, it can be effectively applied to the modeling, analysis, and optimization design of DPW.