翼型(Airfoil)為一種幾何形狀結構，其擁有製造升力的氣動特性，經常應用在飛機機翼、風力渦輪機葉片、直升機螺旋槳、發動機葉片上。目前市面上擁有許多具備良好設計之翼型，但還是有許多產品無法在現有翼型中尋找到適合自己的氣動性能或氣流特性，因此我們期望利用人工智慧的深度學習模型搭配計算流體力學進行翼型設計與開發，目的是降低翼型設計所需的時間與耗費的金錢，讓需要探討翼型繞流場的領域都能夠使用此資源。
本研究透過兩種深度學習模型的串接來進行計算流體力學的加速，實體應用則是以翼型繞流為例。流體力學模擬透過S-A湍流模型來分析與輸出NACA翼型繞流的物理場與空氣動力學係數，並且透過102種不同翼型與0-12度攻角來建立後續深度學習所需之資料集。兩種深度學習方法分別為條件式生成對抗網路模型(Conditional Generative Adversarial Network)與雙重輸入之卷積神經網路模型(Convolutional Neural Network)。條件式生成對抗網路模型(cGAN)能透過幾何與流體力學條件輸入來預測翼型繞流壓力場。卷積神經網路模型(CNN)為雙重輸入之深度學習模型，除了翼型繞流壓力場的輸入外，還額外輸入了流體力學條件，讓模型能夠透過cGAN預測之壓力場圖像與流體力學條件的數值來預測升力係數。兩種深度學習方法的串接能夠快速預測翼型繞流的流體力學性質。能夠提供翼型開發領域一種準確且省時的開發方法。並促進翼型自動優化與設計的研究發展。

An airfoil is a geometrical structure with aerodynamic properties that generate lift. It finds common usage in airplane wings, wind turbine blades, helicopter propellers, and engine blades. While the market offers numerous well-designed airfoils, many products struggle to identify suitable aerodynamic performance or airflow characteristics among existing options. Consequently, we aim to harness artificial intelligence deep learning models in tandem with computational fluid dynamics for airfoil design and development. The objective is to reduce both the time and financial resources expended on airfoil design and to provide this resource for all sectors seeking to explore the realm of airfoil aerodynamics.
In this study, two deep learning models are combined to accelerate computational fluid mechanics. The practical application is focused on the flow around an airfoil. The computational fluid dynamics simulation uses the S-A turbulence model to analyze and output the physical field and aerodynamic coefficients of the airfoil winding. The dataset required for deep learning comprises 102 different airfoils simulated with angles of attack ranging from 0 to 12 degrees. The two deep learning methods employed are the conditional generative adversarial network and the dual input convolutional neural network. The conditional generative adversarial network can predict the pressure field around the airfoil using geometric and CFD conditional inputs. The convolutional neural network is a dual input deep learning model that takes the pressure field around the airfoil and CFD conditions as inputs. This enables the model to predict the lift coefficient based on the pressure field images predicted by the cGAN and the values of the CFD conditions. The integration of these two deep learning models enables rapid prediction of the CFD properties of the airfoils flow. This approach offers an accurate and time-efficient development method within the field of airfoil design, while also advancing research and development in automated airfoil optimization and design.

[1] Z. Han, Z. Gao, W. Song, and L. Xia, “翼型研究的历史、现状与未来发展,” 空气动力学学报, 2021, Vol. 39, Issue 6, Pages: 1-36, vol. 39, no. 6, pp. 1–36, Dec. 2021, doi: 10.7638/KQDLXXB-2021.0396.
[2] P. J. Schubel and R. J. Crossley, “Wind Turbine Blade Design,” Energies 2012, Vol. 5, Pages 3425-3449, vol. 5, no. 9, pp. 3425–3449, Sep. 2012, doi: 10.3390/EN5093425.
[3] P. Fuglsang and C. Bak, “Development of the Risø wind turbine airfoils,” Wind Energy, vol. 7, no. 2, pp. 145–162, Apr. 2004, doi: 10.1002/WE.117.
[4] R. ISAACS, “Airfoil Theory for Rotary Wing Aircraft,” https://doi.org/10.2514/8.11351, vol. 13, no. 4, pp. 218–220, Aug. 2012, doi: 10.2514/8.11351.
[5] S. M. Habali and I. A. Saleh, “Local design, testing and manufacturing of small mixed airfoil wind turbine blades of glass fiber reinforced plastics: Part I: Design of the blade and root,” Energy Convers Manag, vol. 41, no. 3, pp. 249–280, Feb. 2000, doi: 10.1016/S0196-8904(99)00103-X.
[6] S. Daneshmand, C. Aghanajafi, and A. Ahmadi Nadooshan, “The effect of chromium coating in RP technology for airfoil manufacturing,” Sadhana - Academy Proceedings in Engineering Sciences, vol. 35, no. 5, pp. 569–584, Oct. 2010, doi: 10.1007/S12046-010-0036-7/METRICS.
[7] B. Linke, I. Garretson, F. Jan, and M. Hafez, “Integrated Design, Manufacturing and Analysis of Airfoil and Nozzle Shapes in an Undergraduate Course,” Procedia Manuf, vol. 10, pp. 1077–1086, Jan. 2017, doi: 10.1016/J.PROMFG.2017.07.098.
[8] R. Jain, M. Sandeep Jain, M. Lokesh Bajpai, P. G. Student, and P. & Head, “Investigation on 3-D Wing of commercial Aeroplane with Aerofoil NACA 2415 Using CFD Fluent,” International Research Journal of Engineering and Technology, 2016, Accessed: Apr. 17, 2023. [Online]. Available: www.irjet.net
[9] S. Goel, J. I. Cofer, and H. Singh, “Turbine Airfoil Design Optimization,” ASME 1996 International Gas Turbine and Aeroengine Congress and Exhibition, GT 1996, vol. 1, Feb. 2015, doi: 10.1115/96-GT-158.
[10] V. Parezanovic, B. Rasuo, M. Mirko Adzic, V. Parezanovic, B. Rasuo, and M. Adzic, “DESIGN OF AIRFOILS FOR WIND TURBINE BLADES Design Optimization of a Sweeping Jet (SWJ) Actuator for Active Flow Control View project DESIGN OF AIRFOILS FOR WIND TURBINE BLADES,” 2005, Accessed: Apr. 17, 2023. [Online]. Available: https://www.researchgate.net/publication/228608628
[11] C. Bak, P. Fuglsang, N. N. Soerensen, H. Aagaard Madsen, W. Z. Shen, and J. Noerkaer Soerensen, “Airfoil characteristics for wind turbines.” 1999.
[12] O. Günel, E. Koç, and T. Yavuz, “CFD vs. XFOIL of airfoil analysis at low reynolds numbers,” 2016 IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2016, pp. 628–632, 2016, doi: 10.1109/ICRERA.2016.7884411.
[13] K. O’Shea and R. Nash, “An Introduction to Convolutional Neural Networks,” Int J Res Appl Sci Eng Technol, vol. 10, no. 12, pp. 943–947, Nov. 2015, doi: 10.22214/ijraset.2022.47789.
[14] J. Gu et al., “Recent advances in convolutional neural networks,” Pattern Recognit, vol. 77, pp. 354–377, May 2018, doi: 10.1016/J.PATCOG.2017.10.013.
[15] L. O. Chua and T. Roska, “The CNN Paradigm,” IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications, vol. 40, no. 3, pp. 147–156, 1993, doi: 10.1109/81.222795.
[16] H. Chen, L. He, W. Qian, and S. Wang, “Multiple Aerodynamic Coefficient Prediction of Airfoils Using a Convolutional Neural Network,” Symmetry 2020, Vol. 12, Page 544, vol. 12, no. 4, p. 544, Apr. 2020, doi: 10.3390/SYM12040544.
[17] V. Sekar, Q. Jiang, C. Shu, and B. C. Khoo, “Fast flow field prediction over airfoils using deep learning approach,” Physics of Fluids, vol. 31, no. 5, p. 057103, May 2019, doi: 10.1063/1.5094943.
[18] H. WU, X. LIU, W. AN, and H. LYU, “A generative deep learning framework for airfoil flow field prediction with sparse data,” Chinese Journal of Aeronautics, vol. 35, no. 1, pp. 470–484, Jan. 2022, doi: 10.1016/J.CJA.2021.02.012.
[19] W. Chen and A. Ramamurthy, “Deep generative model for efficient 3d airfoil parameterization and generation,” AIAA Scitech 2021 Forum, pp. 1–16, 2021, doi: 10.2514/6.2021-1690.
[20] Q. Du, T. Liu, L. Yang, L. Li, D. Zhang, and Y. Xie, “Airfoil design and surrogate modeling for performance prediction based on deep learning method,” Physics of Fluids, vol. 34, no. 1, p. 015111, Jan. 2022, doi: 10.1063/5.0075784.
[21] B. Deignan, E. Harvey, and L. Hoffman-Goetz, “Fright factors about wind turbines and health in Ontario newspapers before and after the Green Energy Act,” http://dx.doi.org/10.1080/13698575.2013.776015, vol. 15, no. 3, pp. 234–250, May 2013, doi: 10.1080/13698575.2013.776015.
[22] S. Hao et al., “A circular economy approach to green energy: Wind turbine, waste, and material recovery,” Science of The Total Environment, vol. 702, p. 135054, Feb. 2020, doi: 10.1016/J.SCITOTENV.2019.135054.
[23] A. Midilli, I. Dincer, and M. Ay, “Green energy strategies for sustainable development,” Energy Policy, vol. 34, no. 18, pp. 3623–3633, Dec. 2006, doi: 10.1016/J.ENPOL.2005.08.003.
[24] Y. Tang, Y. Zhang, A. Hasankhani, and J. Vanzwieten, “Adaptive Super-Twisting Sliding Mode Control for Ocean Current Turbine-Driven Permanent Magnet Synchronous Generator,” Proceedings of the American Control Conference, vol. 2020-July, pp. 211–217, Jul. 2020, doi: 10.23919/ACC45564.2020.9147480.
[25] S. H. Lee, S. H. Lee, K. Jang, J. Lee, and N. Hur, “A numerical study for the optimal arrangement of ocean current turbine generators in the ocean current power parks,” Current Applied Physics, vol. 10, no. 2, pp. S137–S141, Mar. 2010, doi: 10.1016/J.CAP.2009.11.018.
[26] F. C. Campbell, “Introduction to Composite Materials,” Structural Composite Materials, pp. 1–29, Nov. 2010, doi: 10.31399/ASM.TB.SCM.T52870001.
[27] J. E. Matsson, J. A. Voth, C. A. McCain, and C. McGraw, “Aerodynamic performance of the NACA 2412 Airfoil at low reynolds number,” ASEE Annual Conference and Exposition, Conference Proceedings, vol. 2016-June, Jun. 2016, doi: 10.18260/P.26539.
[28] Z. Yang, C. H. Yu, and M. J. Buehler, “Deep learning model to predict complex stress and strain fields in hierarchical composites,” Sci Adv, vol. 7, no. 15, Apr. 2021, doi: 10.1126/SCIADV.ABD7416/SUPPL_FILE/ABD7416_SM.PDF.
[29] M.-K. Hsu et al., “A Deep Learning Empowered Smart Representative Volume Element Method for Long Fiber Woven Composites,” Front Mater, vol. 10, p. 206, doi: 10.3389/FMATS.2023.1179710.
[30] M. Mirza and S. Osindero, “Conditional Generative Adversarial Nets,” Nov. 2014, Accessed: Apr. 25, 2023. [Online]. Available: https://arxiv.org/abs/1411.1784v1
[31] C.-J. Huang, Y.-H. Chen, Y. Ma, and P.-H. Kuo, “Multiple-Input Deep Convolutional Neural Network Model for COVID-19 Forecasting in China,” medRxiv, p. 2020.03.23.20041608, Mar. 2020, doi: 10.1101/2020.03.23.20041608.
[32] Y. Sun, L. Zhu, G. Wang, and F. Zhao, “Multi-Input Convolutional Neural Network for Flower Grading,” Journal of Electrical and Computer Engineering, vol. 2017, 2017, doi: 10.1155/2017/9240407.
[33] E. Yilmaz and B. J. German, “A convolutional neural network approach to training predictors for airfoil performance,” 18th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, 2017, 2017, doi: 10.2514/6.2017-3660.
[34] F. Moukalled, L. Mangani, and M. Darwish, “The finite volume method,” Fluid Mechanics and its Applications, vol. 113, pp. 103–135, 2016, doi: 10.1007/978-3-319-16874-6_5/FIGURES/22.
[35] Y. You, S. Wang, W. Lv, Y. Chen, and U. Gross, “A CFD model of frost formation based on dynamic meshes technique via secondary development of ANSYS fluent,” Int J Heat Fluid Flow, vol. 89, p. 108807, Jun. 2021, doi: 10.1016/J.IJHEATFLUIDFLOW.2021.108807.
[36] J. E. Matsson, J. A. Voth, C. A. McCain, and C. McGraw, “Aerodynamic performance of the NACA 2412 Airfoil at low reynolds number,” ASEE Annual Conference and Exposition, Conference Proceedings, vol. 2016-June, Jun. 2016, doi: 10.18260/P.26539.
[37] S. R. Allmaras, F. T. Johnson, and P. R. Spalart, “Modications and Clarications for the Implementation of the Spalart-Allmaras Turbulence Model * YourEncore; Boeing Commercial Airplanes retired * * Senior Technical Fellow, Boeing Commercial Airplanes,” pp. 9–13, 2012.
[38] X. Liang, Z. Hu, H. Zhang, C. Gan, and E. P. Xing, “Recurrent Topic-Transition GAN for Visual Paragraph Generation.” pp. 3362–3371, 2017.
[39] O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9351, pp. 234–241, 2015, doi: 10.1007/978-3-319-24574-4_28/COVER.
[40] L. O. Chua and T. Roska, “The CNN Paradigm,” IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications, vol. 40, no. 3, pp. 147–156, 1993, doi: 10.1109/81.222795.
[41] U. Sara, M. Akter, M. S. Uddin, U. Sara, M. Akter, and M. S. Uddin, “Image Quality Assessment through FSIM, SSIM, MSE and PSNR—A Comparative Study,” Journal of Computer and Communications, vol. 7, no. 3, pp. 8–18, Mar. 2019, doi: 10.4236/JCC.2019.73002.
[42] M. R. Veall and K. F. Zimmermann, “PSEUDO-R2 MEASURES FOR SOME COMMON LIMITED DEPENDENT VARIABLE MODELS,” J Econ Surv, vol. 10, no. 3, pp. 241–259, Sep. 1996, doi: 10.1111/J.1467-6419.1996.TB00013.X.