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.
[1]MCCROSKEY W J. A critical assessment of wind tunnel results for the NACA 0012 airfoil: NASA-TM-88302[R]. Washington, D.C.: NASA, 1987.
[2]张伟伟, 王旭, 寇家庆. 面向流体力学的多范式融合研究展望[J]. 力学进展, 2023, 53(2): 433-467.
ZHANG W W, WANG X, KOU J Q. Research prospects of multi-paradigm fusion for fluid mechanics[J]. Advances in Mechanics, 2023, 53(2): 433-467 (in Chinese).
[3]唐志共, 朱林阳, 向星皓, 等. 智能空气动力学若干研究进展及展望[J]. 空气动力学学报, 2023, 41(7): 1-35.
TANG Z G, ZHU L Y, XIANG X H, et al. Progress and prospects of intelligent aerodynamics[J]. Acta Aerodynamica Sinica, 2023, 41(7): 1-35 (in Chinese).
[4]DURU C, ALEMDAR H, BARAN O U. A deep learning approach for the transonic flow field predictions around airfoils[J]. Computers & Fluids, 2022, 236: 105312.
[5]TEJERO F, SURESHBABU S, BOSCAGLI L, et al. Point-enhanced convolutional neural network: a novel deep learning method for transonic wall-bounded flows[J]. Aerospace Science and Technology, 2024, 155: 109689.
[6]DENG Z, WANG J, LIU H, et al. Prediction of transonic flow over supercritical airfoils using geometric-encoding and deep-learning strategies[J]. Aerospace Science and Technology, 2024, 145: 108850.
[7]陈泽伟, 李立, 孔轶男, 等. 基于深度神经网络的高速翼型流场降阶模型[J]. 空气动力学学报, 2025, 43(10): 33-43.
CHEN Z W, LI L, KONG Y N, et al. Reduced-order model for high-speed airfoil flow field based on deep neural network[J]. Acta Aerodynamica Sinica, 2025, 43(10): 33-43 (in Chinese).
[8]CATALANI G, COSTERO D, BAUERHEIM M, et al. A comparative study of learning techniques for the compressible aerodynamics over a transonic RAE2822 airfoil[J]. Computers & Fluids, 2023, 251: 105759.
[9]CHEN L W, THUEREY N. Towards high-accuracy deep learning inference of compressible flows over aerofoils[J]. Computers & Fluids, 2023, 250: 105707.
[10]CATALANI G, AGARWAL S, BERTRAND X, et al. Neural fields for rapid aircraft aerodynamics simulations[J]. Scientific Reports, 2024, 14: 25496.
[11]赵书乐, 张伟伟. 基于表面无黏流动特征的摩阻分布机器学习[J]. 力学学报, 2024, 56(8): 2243-2258.
ZHAO S L, ZHANG W W. Machine learning of skin-friction distribution based on surface inviscid flow features[J]. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(8): 2243-2258 (in Chinese).
[12]WANG H, CAO Y, HUANG Z, et al. Recent advances on machine learning for computational fluid dynamics: a survey[EB/OL]. (2024-08-12) [2026-04-16].
https://arxiv.org/abs/2408.12171.
[13]NABIAN M A, CHOUDHRY S. A mixture of experts gating network for enhanced surrogate modeling in external aerodynamics[EB/OL]. (2025-08-21)[2026-04-16]. https://arxiv.org/abs/2508.21249.
[14]RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707.
[15]PATEL R G, MANICKAM I, TRASK N A, et al. Thermodynamically consistent physics-informed neural networks for hyperbolic systems[J]. Journal of Computational Physics, 2022, 449: 110754.
[16]HARMENING J H, STILLER J, KLEIN L, et al. Data-assisted training of a physics-informed neural network to predict the separated Reynolds-averaged turbulent flow field around an airfoil under variable angles of attack[J]. Neural Computing and Applications, 2024, 36(26): 17267-17282.
[17]WASSING S, LANGER S, BEKEMEYER P. Physics-informed neural networks for transonic flows around an airfoil[EB/OL]. (2024-08-17)[2026-04-16].
https://arxiv.org/abs/2408.17364.
[18]LADSON C L, HILL A S, JOHNSON W G. Pressure distributions from high Reynolds number transonic tests of an NACA 0012 airfoil in the Langley 0.3-meter transonic cryogenic tunnel: NASA-TM-100526[R]. Hampton: NASA Langley Research Center, 1987.
[19]张好, 沈洋, 黄伟, 等. 飞行器智能流场建模方法研究进展[J]. 国防科技大学学报, 2026, 48(1): 1-15.
ZHANG H, SHEN Y, HUANG W, et al. Research progress on intelligent flow field modeling method for aircraft[J]. Journal of National University of Defense Technology, 2026, 48(1): 1-15 (in Chinese).
[20]谷润平, 宋国萍, 刘薇. 高雷诺数下二维翼型绕流气动特性数值分析[J]. 科学技术与工程, 2014, 14(21): 162-167.
GU R P, SONG G P, LIU W. Numerical analysis of aerodynamic characteristics of two-dimensional airfoil flow at high Reynolds number[J]. Science Technology and Engineering, 2014, 14(21): 162-167 (in Chinese).
[21]万芸怡, 黄锐, 刘豪杰. 基于数据驱动的变体机翼跨声速颤振分析[J]. 力学学报, 2025, 57(2): 523-534.
WAN Y Y, HUANG R, LIU H J. Transonic flutter analysis of a morphing wing via data driven method[J]. Chinese Journal of Theoretical and Applied Mechanics, 2025, 57(2): 523-534 (in Chinese).
[22]奕建苗, 邓枫, 覃宁, 等. 快速预测跨声速流场的深度学习方法[J]. 航空学报, 2022, 43(11): 526747.
YI J M, DENG F, QIN N, et al. Fast prediction of transonic flow field using deep learning method[J]. Acta Aeronautica et Astronautica Sinica, 2022, 43(11): 526747 (in Chinese).
[23]PANG Z, LIU K, XIAO H, et al. A deep-learning super-resolution reconstruction model of turbulent reacting flow[J]. Computers & Fluids, 2024, 275: 106249.
CJA