Research on aerodynamic shape optimization of reentry vehicle based on hybrid scale multi-fidelity neural network model

Abstract

The mechanical expanded reentry vehicle has gained significant attention as a reliable solution for shuttle transportation and deep space exploration missions. However, the aerodynamic characteristics of such a reentry vehicle, including high dimensionality, strong nonlinearity, and parameter coupling, pose a challenge in achieving a balance between precision and efficiency in aerodynamic shape design. In this paper, we propose a hybrid scale multi-fidelity neural network (HS-MFNN) model by integrating low-fidelity and high-fidelity models with neural networks. This model is applied to optimize the aerodynamic shape of the reentry vehicle, addressing the challenge. The feasibility of applying the HS-MFNN model was validated through testing using the MBR function. The advantages of the HS-MFNN model were highlighted through a comparison with other models. Subsequently, the model was utilized in the aerodynamic optimization process of the reentry vehicle. The prediction error for the drag coefficient (Cd) was less than 1%, and for the head stagnation heat flux (QO), it was within 6%. Moreover, the computation time was reduced by three orders of magnitude, ensuring both computational accuracy and efficiency. After optimization, the Cd increased by 5.08%, while the QO decreased by 8.62%. These improvements demonstrate that the use of the HS-MFNN model significantly enhances the aerodynamic performance of the reentry vehicle. Consequently, the HS-MFNN model exhibits great potential for fast and efficient optimization processes.