关键词: 旋翼翼型, 宽速域流场预测, 门控混合专家, 物理约束, 跨声速流动

Abstract: To address the challenges in wide-speed-range flow-field prediction for rotor airfoils, including the coexistence of multiple flow regimes, pronounced shock-structure migration, and the difficulty of conventional single-network architectures in simultaneously capturing global accuracy and local strong-gradient features, this paper proposes a Gated Mixture-of-Experts Deep Residual Network (MoE-ResNet) method for rapid flow-field prediction of rotor airfoils over a wide speed range. Built upon a residual neural network backbone, the proposed method incorporates a gated mixture-of-experts residual architecture to enhance the model’s adaptive representation capability for multimodal flow features under wide operating conditions. In addition, segmented physical constraints and an adaptive weighting training strategy are introduced to improve learning stability and physical plausibility in transonic strong-gradient regions. The effectiveness of the proposed method is validated through comparison with NASA wind-tunnel pressure-coefficient (Cp) distributions, and the generalization performance of the model is assessed using a rigorous blind-test protocol. Results show that, compared with the artificial-viscosity-based PINN method, the proposed method reduces the maximum absolute error of the surface pressure coefficient by up to 89.5%. Under all blind-test conditions, the mean relative L2 error of the pressure field is 4.214%, the mean relative error of the lift coefficient is 3.420%, the mean shock-position error under transonic conditions is 0.0042c, and the mean absolute error of surface pressure-coefficient prediction ranges from 0.003 to 0.013. In addition, the proposed method achieves a 338-fold speedup over CFD. While maintaining physical plausibility, the proposed method demonstrates high-accuracy prediction capability for both the global flow field and local strong-gradient regions of rotor airfoils over a wide speed range.

Key words: rotor airfoil, wide-speed-range flow-field prediction, gated mixture-of-experts, physical constraints, transonic flow