在現代無人戰鬥飛行載具（Unmanned Combat Aerial Vehicle, UCAV）的氣動外型設計過程中，其氣動性能的評估高度依賴計算流體力學（Computational Fluid Dynamics, CFD）模擬，然而其高昂的計算負載與冗長的求解週期嚴重限制設計迭代效率，此外外型設計流程高度仰賴工程經驗與複雜與反覆試錯，顯著提高研發成本與門檻。
近年來隨著深度學習（Deep Learning）技術的興起為加速CFD預測提供創新路徑。有別於傳統氣動外型設計需依賴複雜的外型參數化及人為設定之幾何限制，生成式類神經網路（Generative Neural Networks, GNN）展現數據驅動設計的優勢，透過從既有資料庫中提取潛在向量（Latent code）並將其映射至低維空間，GNN能有效重構並生成具備物理特徵的新穎外型設計。
本研究選用NATO AVT-161計畫提出之SACCON構型作為UCAV基準構型，透過變化其後掠角與多樣化翼型配置，構建涵蓋廣泛亞音速流場特性的UCAV幾何與氣動資料庫。在設計生成端，利用變分自編碼器（VAE）生成式模型，實現從低維潛在空間重構三維幾何外型，有效避開傳統參數化建模的複雜性與限制；在性能評估端，訓練完成後的CFD代理模型在維持預測精度的前提下，顯著降低候選外型之評估成本。
本研究將生成式幾何重構與快速氣動預測技術應用於 UCAV 外型設計，完成 VAE 幾何生成模型與 CFD 代理模型之初步工具串接，並驗證候選幾何由潛在空間生成後，可進一步轉換為表面壓力分佈預測結果。現階段研究重點在於確認資料表徵、候選幾何生成與快速氣動評估之間的流程可行性，尚未納入明確的氣動目標函數、設計限制條件、最佳化搜尋策略及高保真 CFD 回驗證，因此本研究成果應定位為完整氣動外型優化方法框架之前期方法論與工具基礎，而非已完成之自動化外型最佳化流程。

This study develops a data-driven framework for unmanned combat aerial vehicle (UCAV) aerodynamic shape design by combining a computational fluid dynamics (CFD) surrogate model with a variational autoencoder (VAE)-based geometry generation model. In conventional aerodynamic design, CFD simulations are widely used to evaluate candidate geometries. Although CFD provides high-fidelity aerodynamic information, its high computational cost limits large-scale design exploration and iterative optimization. In addition, three-dimensional UCAV configurations involve complex and highly coupled geometric variations, making it difficult to construct an efficient design space using only manually defined parameters.
To address these issues, this study adopts the SACCON / F17E configuration as the baseline UCAV geometry. A database is constructed by varying sweep angle, airfoil profile, Mach number, and angle of attack. The three-dimensional geometries and CFD results are converted into two-dimensional tensor representations, including thickness maps, camber maps, and upper and lower surface pressure distributions. This unified representation enables the CFD surrogate model and the VAE geometry generation model to operate within a consistent data format.
For aerodynamic prediction, a UNet-based CFD surrogate model is developed to predict surface pressure distributions from geometry images and flow conditions. Different strategies for incorporating flow conditions are investigated, including input-level condition representation and feature-level condition fusion. The results show that the input-level condition representation model provides stable and accurate pressure-field predictions under the present dataset and flow-condition settings, and is therefore selected for the subsequent coupling process.
For geometry generation, a VAE model is trained to learn a low-dimensional latent representation of UCAV geometries. The interpolation and sampling results show that the VAE can reconstruct geometry images and generate candidate shapes with continuous variations in sweep angle, thickness distribution, and camber distribution. Finally, the generated geometries are evaluated by the CFD surrogate model under specified flow conditions, demonstrating a preliminary workflow that connects latent-space geometry generation with rapid aerodynamic prediction.
The present study does not yet include a complete optimization algorithm or high fidelity CFD validation of optimized candidates. Instead, it establishes the fundamental methodology for integrating data-driven geometry generation and surrogate-based aerodynamic evaluation, providing a basis for future UCAV aerodynamic shape optimization.

[1] R. Cummings and A. Schütte, "An integrated computational/experimental approach to UCAV stability & control estimation: overview of NATO RTO AVT-161," 28th AIAA Applied Aerodynamics Conference, p. 4392, 2010.
[2] T. Loeser, D. Vicroy, and A. Schuette, "SACCON static wind tunnel tests at DNW-NWB and 14 x22 NASA LaRC," 28th AIAA applied aerodynamics conference, p. 4393, 2010.
[3] A. Schütte, D. Hummel, and S. M. Hitzel, "Numerical and experimental analyses of the vortical flow around the SACCON configuration," 28th AIAA Applied Aerodynamics Conference, p. 4690, 2010.
[4] D. M. Nelson, J. P. Irving, M. Ghoreyshi, A. Jirasek, and A. J. Lofthouse, "Experimental and numerical investigation of flight dynamics of a generic lambda wing configuration," Aerospace Science and Technology, vol. 71, pp. 706–724, 2017.
[5] M. Rein, J. Irving, G. Rigby, and T. J. Birch, "High speed static experimental investigations to estimate control device effectiveness and S&C capabilities," 32nd AIAA Applied Aerodynamics Conference, p. 2004, 2014.
[6] N. T. Frink, M. Tormalm, and S. Schmidt, "Unstructured CFD aerodynamic analysis of a generic UCAV configuration," NATO RTO AVT-189 Specialist Meeting, no. RTO-MP-AVT-170, 2011.
[7] M. Norgaard, C. C. Jorgensen, and J. C. Ross, "Neural network prediction of new aircraft design coefficients," 1997.
[8] Y. Zhang, W. J. Sung, and D. N. Mavris, "Application of convolutional neural network to predict airfoil lift coefficient," 2018 AIAA/ASCE/AHS/ASC structures, structural dynamics, and materials conference, p. 1903, 2018.
[9] N. Thuerey, K. Weißenow, L. Prantl, and X. Hu, "Deep learning methods for Reynolds-averaged Navier–Stokes simulations of airfoil flows," AIAA journal, vol. 58, no. 1, pp. 25–36, 2020.
[10] 游博元, "基於深度學習之三維凸起物於超音速流場性能預測," 國立成功大學碩士學位論文, 2023.
[11] W. Haizhou, L. Xuejun, A. Wei, and L. Hongqiang, "A generative deep learning framework for airfoil flow field prediction with sparse data," Chinese Journal of Aeronautics, vol. 35, no. 1, pp. 470–484, 2022.
[12] L. Xiao, M. Zhang, and X. Chang, "Learning airfoil flow field representation via geometric attention neural field," Applied Sciences, vol. 14, no. 22, p. 10685, 2024.
[13] U. Michelucci, "An introduction to autoencoders," arXiv preprint arXiv:2201.03898, 2022.
[14] D. P. Kingma and M. Welling, "Auto-encoding variational bayes," arXiv preprint arXiv:1312.6114, 2013.
[15] S. Bowman, L. Vilnis, O. Vinyals, A. Dai, R. Jozefowicz, and S. Bengio, "Generating sentences from a continuous space," Proceedings of the 20th SIGNLL conference on computational natural language learning, pp. 10–21, 2016.
[16] I. Higgins et al., "beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework," International Conference on Learning Representations, 2016.
[17] I. J. Goodfellow et al., "Generative adversarial nets," Advances in neural information processing systems, vol. 27, 2014.
[18] L. Regenwetter, A. H. Nobari, and F. Ahmed, "Deep generative models in engineering design: A review," Journal of Mechanical Design, vol. 144, no. 7, p. 071704, 2022.
[19] D. D'Agostino, A. Serani, E. F. Campana, and M. Diez, "Deep autoencoder for off-line design-space dimensionality reduction in shape optimization," 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, p. 1648, 2018.
[20] W. Chen, K. Chiu, and M. D. Fuge, "Airfoil design parameterization and optimization using bézier generative adversarial networks," AIAA journal, vol. 58, no. 11, pp. 4723–4735, 2020.
[21] X. Du, P. He, and J. R. Martins, "A B-spline-based generative adversarial network model for fast interactive airfoil aerodynamic optimization," AIAA scitech 2020 forum, p. 2128, 2020.
[22] W. Chen and A. Ramamurthy, "Deep generative model for efficient 3D airfoil parameterization and generation," AIAA Scitech 2021 Forum, p. 1690, 2021.
[23] Y. Wang, K. Shimada, and A. Barati Farimani, "Airfoil GAN: Encoding and synthesizing airfoils for aerodynamic shape optimization," Journal of Computational Design and Engineering, vol. 10, no. 4, pp. 1350–1362, 2023.
[24] J. Jabón, S. Corbera, R. Álvarez, and R. Barea, "Aerodynamic shape optimization using graph variational autoencoders and genetic algorithms," Structural and Multidisciplinary Optimization, vol. 67, no. 3, p. 32, 2024.
[25] Y.-E. Kang, D. Lee, and K. Yee, "Intuitive and feasible geometric representation of airfoil using variational autoencoder," Journal of Computational Design and Engineering, vol. 12, no. 2, pp. 27–48, 2025.
[26] L. F. G. Marcantoni, J. P. Tamagno, and S. A. Elaskar, "High speed flow simulation using openfoam," Mecánica Computacional, vol. 31, no. 16, pp. 2939–2959, 2012.