The fusion method based on small-sample aerodynamic thermal and force data.

This article presents a deep neural network (DNN)-based semi-supervised learning method for fusing small-sample high-fidelity aerodynamic and aerothermal data with abundant low-fidelity computational fluid dynamics (CFD) data. The approach introduces unlabeled low-fidelity data as soft constraints via a dynamically weighted loss function to mitigate overfitting and label noise, thereby enhancing prediction accuracy under limited high-fidelity data conditions. Validation is performed using surface pressure distribution data from the ONERA (National Office for Aerospace Studies and Research) M6 wing and aerodynamic heating data from a blunt bicone vehicle, demonstrating superior performance compared to traditional single-fidelity and existing multi-fidelity models, especially as the amount of high-fidelity data decreases. The method shows promise for efficient and accurate aerodynamic thermal load prediction in aircraft design and development, with future work aimed at improving robustness and integrating hybrid modeling techniques.