極音速武器早在二戰時已提出構想，近期由於俄羅斯在入侵烏克蘭期間使用極音速飛彈，使得世界大國更加重視發展極音速載具，而飛行載具於極音速的環境下也將面臨更為苛刻的飛行條件，因此載具表面須承擔更高的阻力與熱負載。透過將極音速載具鼻尖處(即鈍體前緣)前方增加尖釘被視為一種簡單且容易降低阻力以及熱負載的方法，許多研究都使用數值模擬方式來進行阻力與熱負載的評估，然而，利用傳統的數值模擬手段來設計極音速載具的外型會消耗大量的時間。故本研究採用卷積神經網路模型，針對含有尖釘之鈍體進行流場的預測，這不僅省去傳統的數值模擬所消耗之大量的時間和資源，並有助於設計極音速載具之外型與熱防護材料之快速預測。
本研究運用數值模擬5馬赫的極音速飛彈於飛行高度17公里狀態下的熱傳行為。透過CFD對具有尖釘之鈍體進行數值模擬，將物理場數值(包含馬赫場、溫度場、壓力場與鈍體表面之空氣動力熱通量)透過資料點採樣輸出成U-net 卷積神經網路模型所需的數據圖並彙集成資料庫。本研究建立一卷積神經網路模型並採用U-Net結構來建立預測模型。透過符號距離函數、物理場數據圖及流域幾何形狀之像素圖輸入至U-net CNN模型進行訓練，其結果透過三個通道輸出，並將其模型與權重保存透過匯入不同尖釘之鈍體的外型像素圖於保存的U-net CNN模型進行預測，可以快速預測尖釘之鈍體周圍的流場並準確捕獲流場的流動結構。U-net CNN模型的整體準確率為0.875，而經過優化的卷積神經網路，可以使整體準確率提升至0.9365，加速提升約95.46%。針對具有尖釘之鈍體的熱通量預測上，其時間跟優化過後的卷積神經網路耗時無明顯差異，而整體準確率為0.85，在預測資料庫以外具有尖釘之鈍體的熱通量分布圖預測結果上，最大熱通量誤差約22.1%。

Russia's use of hypersonic missiles during the invasion of Ukraine has spurred major world powers to focus on developing hypersonic missiles/vehicles. These missiles/vehicles face severe conditions, requiring their surfaces to withstand high drag and thermal load. Adding a spike in front of the blunt body nose is a simple method to reduce drag and thermal load. Traditional numerical simulations for designing hypersonic missiles/vehicles are time-consuming. This study employs a Convolutional Neural Network (CNN) model to predict the flow field around a blunt body with a spike, saving time and resources and aiding in rapid exterior design of missile/vehicle and the shape design of thermal protection material.
The study simulates the aerodynamic heating behavior of a hypersonic missile at Mach 5 and 17 km altitude. CFD is utilized to simulate the hypersonic flow over the spiked blunt body, and then physical quantity (i.e., Mach, temperature, pressure) and gradient of physical quantity (e.g., aerodynamic heat flux) are collected for fluid domain and the solid domain by given coordinates as the image data of database for U-net CNN model. The U-net CNN model was trained with signed distance functions, physical field data, and pixel images of flow domain geometry. Results were output through three channels, and the U-net CNN model with weights was saved. By inputting different blunt body geometries with spikes, the U-net CNN model can quickly and accurately predict the flow field and capture flow structures. The U-net CNN model's overall accuracy was 0.875, and optimization increased accuracy to 0.9365, with a speed improvement of approximately 95.46%. For heat flux prediction of the blunt body with a spike, the accuracy was 0.85, with the maximum error of about 22.1%.

【1】 Mach, Ernst, and Peter Salcher. Photographische Fixirung der durch Projectile in der Luft eingeleiteten Vorgänge. Annalen der Physik, 277-291. 1887.
【2】 Leishman, J. Gordon. Introduction to Aerospace Flight Vehicles. Embry-Riddle Aeronautical University Publishing .2023.
【3】 Keyes, J. Wayne, and Frank D. Hains. Analytical and experimental studies of shock interference heating in hypersonic flows .No. L-8423.NASA. 1973.
【4】 Wieting, Allan R., and Michael S. Holden. Experimental shock-wave interference heating on a cylinder at Mach 6 and 8. AIAA Journal 27(11),1557-1565. 1989.
【5】 Menter, Florian R. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA Journal 32.8, 1598-1605. 1994.
【6】 Ye, K., Ye, Z., Li, C., & Wu, J. Effects of the aerothermoelastic deformation on the performance of the three-dimensional hypersonic inlet. Aerospace Science and Technology, 84, 747-762. 2019.
【7】 Deng, X., Mao, M., Tu, G., Zhang, H., & Zhang, Y. High-order and high accurate CFD methods and their applications for complex grid problems. Communications in Computational Physics, 11(4), 1081-1102. 2012.
【8】 Von Kármán, T. Aerodynamics: selected topics in the light of their historical development. Courier Corporation. 2004.
【9】 Bérend, N. The Missing Calculation behind the Original Kármán Line Definition-A Credible Hypothesis. IAC 2021. Oct 2021.
【10】 Bogdonoff, Seymour M., and Irwin E. Vas. Preliminary investigations of spiked bodies at hypersonic speeds. Journal of the Aerospace Sciences, 26(2), 65-74. 1959.
【11】 Allen, H. Julian, and Alfred J. Eggers Jr. A study of the motion and aerodynamic heating of missiles entering the earth's atmosphere at high supersonic speeds .No. NACA-RM-A53D28.NASA. 1953.
【12】 Motoyama, Noboru, et al. Thermal protection and drag reduction with use of spike in hypersonic flow. 10th AIAA NAL-NASDA-ISAS International Space Planes and Hypersonic Systems and Technologies Conference. 2001.
【13】 Crawford, Davis H. Investigation of the flow over a spiked-nose hemisphere-cylinder at a Mach number of 6.8. National Aeronautics and Space Administration. 1959.
【14】 黃柏勳。具有尖釘之鈍體於極音速流場環境下氣動加熱行為之數值模擬分析。碩士學位論文。國立成功大學。2023。
【15】 Qin, Qihao, Jinglei Xu, and Shuai Guo. Fluid-thermal analysis of aerodynamic heating over spiked blunt body configurations. Acta Astronautica, 132, 230-242. 2017.
【16】 Schaffer, Simon. Babbage's intelligence: Calculating engines and the factory system. Critical Inquiry, 21(1), 203-227. 1994.
【17】 Bengio, Y. Learning deep architectures for AI. Foundations and trends® in Machine Learning, 2(1), 1-127. 2009.
【18】 LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521(7553), 436-444. 2015.
【19】 O'shea, Keiron, and Ryan Nash. An introduction to convolutional neural networks. arXiv preprint arXiv:1511.08458. 2015.
【20】 Guo, Xiaoxiao, Wei Li, and Francesco Iorio. Convolutional neural networks for steady flow approximation. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining (pp. 481-490). 2016.
【21】 Rojek, Krzysztof, Roman Wyrzykowski, and Pawel Gepner. AI-accelerated CFD simulation based on OpenFOAM and CPU/GPU computing. In Computational Science–ICCS 2021: 21st International Conference, Krakow, Poland, June 16–18, 2021, Proceedings, Part II 21 (pp. 373-385). Springer International Publishing. 2021.
【22】 Potluri, S., Fasih, A., Vutukuru, L. K., Al Machot, F., & Kyamakya, K. CNN based high performance computing for real time image processing on GPU. In Proceedings of the Joint INDS'11 & ISTET'11 (pp. 1-7). IEEE. Jul 2011.
【23】 Ribeiro, M. D., Rehman, A., Ahmed, S., & Dengel, A. DeepCFD: Efficient steady-state laminar flow approximation with deep convolutional neural networks. arXiv preprint arXiv:2004.08826.2020.
【24】 Abucide-Armas, A., Portal-Porras, K., Fernandez-Gamiz, U., Zulueta, E., & Teso-Fz-Betoño, A. (2021). A data augmentation-based technique for deep learning applied to CFD simulations. Mathematics, 9(16), 1843.
【25】 Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 (pp. 234-241). Springer International Publishing. 2015.
【26】 Russo, Giovanni, and Peter Smereka. A remark on computing distance functions. Journal of Computational Physics, 163(1), 51-67. 2000.
【27】 Chicco, Davide, Matthijs J. Warrens, and Giuseppe Jurman. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Computer Science, 7, e623. 2021.
【28】 Fernando, J. R-squared: Definition, calculation formula, uses, and limitations. Investopedia. Retrieved September, 8, 2023.
【29】 Shirzadi, M., Fukasawa, T., Fukui, K., & Ishigami, T. Application of deep learning neural networks for the analysis of fluid-particle dynamics in fibrous filters. Chemical Engineering Journal, 455, 140775. 2023.
【30】 ANSYS Fluent 19.0 theory guide, ANSYS Incorporation. 2019.
【31】 Anaconda Server Documentation Release 6.6.2, Anaconda Incorporation. 2024.
【32】 Learn Matlab2023a CN, The MathWorks Incorporation. 2023.
【33】 Roberts, Jonas. Micro wind energy systems in harsh environments-failure analysis of small wind turbines at remote sites in Labrador, Doctoral dissertation, Memorial University of Newfoundland . 2009.
【34】 Introduction to Introduction to ANSYS FLUENT, ANSYS Incorporation. 2010.
【35】 Kost, Alex. Applying neural networks for tire pressure monitoring systems, Doctoral dissertation, California Polytechnic State University. 2018.
【36】 Portal-Porras, K., Fernandez-Gamiz, U., Ugarte-Anero, A., Zulueta, E., & Zulueta, A. Alternative artificial neural network structures for turbulent flow velocity field prediction. Mathematics, 9(16), 1939. 2021.
【37】 Ahmed, Mahmoud YM, and Ning Qin. Forebody shock control devices for drag and aero-heating reduction: A comprehensive survey with a practical perspective. Progress in Aerospace Sciences, 112, 100585. 2020.