01.1 - Aircraft Design and Integrated Systems (Theory, Methods and Tools)
MULTI-FIDELITY BAYESIAN NEURAL NETWORKS FOR STEALTH/AERODYNAMIC DATA FUSION METHOD
J.Y. He¹, H. Zhao¹; ¹Northwestern Polytechnical University, China
This study introduces a multi-fidelity Bayesian neural network to overcome the accuracy-efficiency trade-off in stealth simulation. By fusing data of varying fidelity and using variational inference for uncertainty estimation, it achieves higher predictive accuracy and more reliable uncertainty quantification than CoKriging, offering an efficient tool for multidisciplinary aircraft optimization.