数据驱动的超临界翼型辅助设计方法研究

超临界翼型的空气动力学问题是商用宽体飞机机翼设计中的关键问题，宽体 飞机的性能直接取决于超临界翼型的空气动力学特性。在过去，解决超临界翼型 空气动力学问题的方法需要求解流场网格每个网格点上复杂的偏微分方程组。然 而，这些具有高维度、非线性和多尺度等的空气动力学问题很难快速得到解析解。 近年来深度学习技术和计算机算力的飞速发展，数据驱动建模的方法利用求解完 毕的大量流场仿真数据并结合深度学习算法，有望高效准确的解决超临界翼型的 空气动力学问题。 由于超临界翼型的气动力性能对其几何外形高度敏感，以及其流场的数据维 度过高的问题，数据很难进行有效的利用。所以本文的主要研究内容为如何有效 的利用数据并通过数据驱动的方式进行超临界翼型的辅助设计。本文的首项工作 为对超临界翼型及其流场数据进行预处理，针对超临界翼型数据，通过非均匀有 理 B 样条（Non-Uniform Rational B-Splines，NURBS）的方法将其几何外形在 10−5 的误差允许范围内拟合为 18 个控制点。针对具有较高维度的压力分布数据通过本 征正交分解（Proper Orthogonal Decomposition，POD）降维为具有 30 个主模态信 息的向量。并通过探索不同形式的数值流场信息，将流场数据处理为神经网络可 输入的形式，即 385×120 的矩形张量。在数据处理的基础上，本文的第二项工作为 超临界翼型性能预测器的建立。通过对多层感知机（Multilayer Perceptron，MLP） 的结构设计，实现了快速、准确的性能预测，最终三种气动力性能预测器的精度都 在 99% 以上。更进一步的，为了实现更详细流场结构与信息的预测，进而达到反 向设计、优化翼型的目标，本文的最后一项工作为基于单个模型设计的翼型-流场 的双向映射器，并且达到了 99% 以上的精度。本文提出的两个模型都具有简洁的 结构以及高精度的结果，可以高效地嵌入超临界翼型的优化设计流程。

[1] OBERT E. Aerodynamic design of transport aircraft[M]. IOS press, 2009.
[2] WHITCOMB R T. Lift and drag characteristics of a wing with several angles of sweep at highsubsonic speeds[C]//NACA. Ames Aeron. Lab. NACA Conf. on Aerodyn. Probl. of TransonicAirplane Design. 1947.
[3] DURIEZ T, BRUNTON S L, NOACK B R. Machine learning control-taming nonlinear dy namics and turbulence[M]. Springer, 2017.
[4] LING J, KURZAWSKI A, TEMPLETON J. Reynolds averaged turbulence modelling usingdeep neural networks with embedded invariance[J]. Journal of Fluid Mechanics, 2016, 807:155-166.
[5] KUTZ J N. Deep learning in fluid dynamics[J]. Journal of fluid mechanics, 2017, 814: 1-4.
[6] BENNER P, GUGERCIN S, WILLCOX K. A survey of projection-based model reductionmethods for parametric dynamical systems[J]. SIAM review, 2015, 57(4): 483-531.
[7] DONAHUE J, ANNE HENDRICKS L, GUADARRAMA S, et al. Long-term recurrent convo lutional networks for visual recognition and description[C]//Proceedings of the IEEE conferenceon computer vision and pattern recognition. 2015: 2625-2634.
[8] ZHANG Y, SUNG W J, MAVRIS D N. Application of convolutional neural network to predictairfoil lift coefficient[C]//2018 AIAA/ASCE/AHS/ASC structures, structural dynamics, andmaterials conference. 2018: 1903.
[9] LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recog nition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[10] MIYANAWALA T P, JAIMAN R K. An efficient deep learning technique for the Navier-Stokesequations: Application to unsteady wake flow dynamics[A]. 2017.
[11] YILMAZ E, GERMAN B. A convolutional neural network approach to training predictors forairfoil performance[C]//18th AIAA/ISSMO multidisciplinary analysis and optimization confer ence. 2017: 3660.
[12] HUI X, BAI J, WANG H, et al. Fast pressure distribution prediction of airfoils using deeplearning[J]. Aerospace science and technology, 2020, 105: 105949.
[13] GUO X, LI W, IORIO F. Convolutional neural networks for steady flow approximation[C]//Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery anddata mining. 2016: 481-490.
[14] YE S, ZHANG Z, SONG X, et al. A flow feature detection method for modeling pressuredistribution around a cylinder in non-uniform flows by using a convolutional neural network[J]. Scientific reports, 2020, 10(1): 1-10.
[15] THUEREY N, WEISSENOW K, PRANTL L, et al. Deep learning methods for Reynolds averaged Navier–Stokes simulations of airfoil flows[J]. AIAA journal, 2020, 58(1): 25-36.
[16] RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional Networks for BiomedicalImage Segmentation[A]. 2015. arXiv: 1505.04597.
[17] SEKAR V, JIANG Q, SHU C, et al. Fast flow field prediction over airfoils using deep learningapproach[J]. Physics of fluids, 2019, 31(5): 057103.
[18] BHATNAGAR S, AFSHAR Y, PAN S, et al. Prediction of aerodynamic flow fields usingconvolutional neural networks[J]. Computational mechanics, 2019, 64(2): 525-545.
[19] TONG Z, ZHANG Y, CHEN H. Multi-objective aerodynamic optimization of supercriticalwing with substantial pressure constraints[C]//53rd AIAA aerospace sciences meeting. 2015:0763.
[20] SEKAR V, ZHANG M, SHU C, et al. Inverse design of airfoil using a deep convolutional neuralnetwork[J]. Aiaa Journal, 2019, 57(3): 993-1003.
[21] GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative Adversarial Networks[A]. 2014. arXiv: 1406.2661.
[22] YILMAZ E, GERMAN B. Conditional generative adversarial network framework for airfoilinverse design[C]//AIAA aviation 2020 forum. 2020.
[23] MIRZA M, OSINDERO S. Conditional Generative Adversarial Nets[A]. 2014. arXiv:1411.1784.
[24] WANG Y Y, ZHANG B Q, CHEN Y C. Robust airfoil optimization based on improved particleswarm optimization method[J]. Applied mathematics and mechanics, 2011, 32.
[25] KENNEDY J, EBERHART R. Particle swarm optimization[C/OL]//Proceedings of ICNN’95- International Conference on Neural Networks: volume 4. 1995: 1942-1948 vol.4. DOI:10.1109/ICNN.1995.488968.
[26] ZIJING L, XUEJUN L, XINYE C. A new hybrid aerodynamic optimization framework basedon differential evolution and invasive weed optimization[J]. Chinese journal of aeronautics,2018, 31(7).
[27] STORN R, PRICE K. Differential evolution-a simple and efficient heuristic for global opti mization over continuous spaces[J]. Journal of global optimization, 1997, 11(4): 341.
[28] LIANG J, QIN A, SUGANTHAN P, et al. Comprehensive learning particle swarm optimizerfor global optimization of multimodal functions[J/OL]. IEEE Transactions on EvolutionaryComputation, 2006, 10(3): 281-295. DOI: 10.1109/TEVC.2005.857610.
[29] BOUHLEL M A, HE S, MARTINS J R. Scalable gradient–enhanced artificial neural networksfor airfoil shape design in the subsonic and transonic regimes[J]. Structural and ultidisciplinaryoptimization, 2020, 61.
[30] WU H, LIU X, AN W, et al. A deep learning approach for efficiently and accurately evaluatingthe flow field of supercritical airfoils[J]. Computers & Fluids, 2020, 198: 104393.
[31] JING W, RUNZE L, CHENG H, et al. An inverse design method for supercritical airfoil basedon conditional generative models[J]. Chinese journal of aeronautics, 2022, 35(3): 62-74.
[32] KINGMA D P, WELLING M. Auto-Encoding Variational Bayes[A]. 2022. arXiv: 1312.6114.
[33] ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein GAN[A]. 2017. arXiv: 1701.07875.
[34] JIN X, CHENG P, CHEN W L, et al. Prediction model of velocity field around circular cylinderover various Reynolds numbers by fusion convolutional neural networks based on pressure onthe cylinder[J]. Physics of Fluids, 2018, 30(4): 047105.
[35] HORNIK K, STINCHCOMBE M, WHITE H. Multilayer feedforward networks are universalapproximators[J]. Neural networks, 1989, 2(5): 359-366.
[36] CYBENKO G. Approximation by superpositions of a sigmoidal function[J]. Mathematics ofcontrol, signals and systems, 1989, 2(4): 303-314.
[37] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convo lutional neural networks[J]. Advances in neural information processing systems, 2012, 25.
[38] ROSENBLATT F. Principles of neurodynamics. perceptrons and the theory of brain mecha nisms[R]. Cornell Aeronautical Lab Inc Buffalo NY, 1961.
[39] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale imagerecognition[A]. 2014.
[40] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]//Proceedings of theIEEE conference on computer vision and pattern recognition. 2015: 1-9.
[41] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedingsof the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[42] MENTER F R. Two-equation eddy-viscosity turbulence models for engineering applications[J]. AIAA journal, 1994, 32(8): 1598-1605.
[43] ROE P L. Approximate Riemann solvers, parameter vectors, and difference schemes[J]. Journalof computational physics, 1981, 43(2): 357-372.
[44] KULFAN B, BUSSOLETTI J. ” Fundamental” parameteric geometry representations for air craft component shapes[C]//11th AIAA/ISSMO multidisciplinary analysis and optimizationconference. 2006.
[45] SOBIECZKY H. Parametric airfoils and wings[J]. Recent development of aerodynamic designmethodologies: inverse design and optimization, 1999.
[46] PIEGL L, TILLER W. The NURBS book[M]. Springer science & Business media, 1996.
[47] CHEN W, CHIU K, FUGE M. Aerodynamic design optimization and shape exploration usinggenerative adversarial networks[C]//AIAA Scitech 2019 Forum. 2019.