計算流體力學(Computational Fluid Dynamics, CFD)透過數值方法離散並求解連續方程式、動量方程式與能量方程式，以獲得密度、速度與壓力等物理量之數值解，進而分析流體運動行為及其與固體邊界之交互作用。其流程包含前處理(幾何建模、網格生成、物理模型與邊界條件設定)、數值求解與後處理等階段。隨著流場問題複雜度提升(如動態網格與高階紊流模型之導入)及預測物理量種類增加，前後處理所需時間與計算資源成本顯著提高。於外型最佳化設計過程中，頻繁之設計變更將反覆產生成本負擔，使快速性能評估成為一項具挑戰性之課題，因此如何降低計算成本並提升設計效率已成為工業界與學術界關注之重要議題。
近年來圖形運算單元(Graphics Processing Unit, GPU)運算能力之提升、深度學習演算法與網路架構之發展，以及開源工具之普及，促使深度學習技術於工程領域快速成長，並在多項應用中展現成效。例如，深度強化學習(Deep Reinforcement Learning, DRL)已成功應用於機器人控制，而卷積神經網路(Convolutional Neural Networks, CNN)則廣泛應用於醫學影像分析與工業缺陷檢測。在空氣動力學領域深度學習可建構 CFD 代理模型(surrogate model)，於數秒內預測模擬結果，提供研究人員作為設計評估依據，顯著加速設計流程。
本研究以次音速三維機翼流場為研究對象，建立涵蓋不同馬赫數(M_{infty })、攻角(α)與前緣後掠角({Lambda }_{LE})之資料庫，資料內容包含升力係數、阻力係數、表面壓力分佈及各截面流場物理量。基於此資料集，本研究對深度神經網路架構進行改良，並導入多任務學習(Multi-task learning, MTL)，同步預測多項氣動任務。透過訓練與驗證程序，評估不同神經網路架構之預測準確度與泛化能力，最終比較各模型於三維機翼流場之預測表現，以驗證其於工程設計加速應用之可行性。

This study investigates the application of Multi-Task Learning (MTL) to subsonic three-dimensional wing flow-field prediction. Single-Task Learning (STL) and MTL architectures were developed for predicting surface pressure distributions, sectional flow fields, and aerodynamic coefficients, and their performances were systematically compared.
The proposed framework employs dual encoders to extract multi-view features and utilizes hard parameter sharing within a shared latent space. Task-specific outputs are generated through independent decoders and multilayer perceptron. Compared with STL models, the MTL framework reduces GPU memory consumption by 35.7% and training time by 28.2%. Although a slight reduction in prediction accuracy is observed, MTL exhibits improved robustness and, under certain test conditions, achieves superior predictive performance.
These results demonstrate that effective knowledge sharing can be achieved among aerodynamic tasks and that the proposed MTL framework provides a favorable balance between accuracy, robustness, and computational efficiency, highlighting its potential for aerodynamic analysis and engineering design optimization.

1. S. Ruder, “An Overview of Multi-Task Learning in Deep Neural Networks”, arXiv preprint arXiv:1706.05098, 2017.
2. M. T. Rosenstein, Z. Marx, L. P. Kaelbling and T. G. Dietterich, “To Transfer or Not to Transfer”, in Proceedings of the Inductive Transfer: 10 Years Later NIPS 2005 Workshop, 2005.
3. OpenFOAM Ltd, “OpenFOAM: Tutorial Guide”, [Online]. Available: https://www.openfoam.com/documentation/tutorial-guide (Accessed: Jun. 1, 2026), 2017.
4. J. A. Heyns, O. F. Oxtoby, and A. Steenkamp, “Modelling High-Speed Flow Using a Matrix-Free Coupled Solver”, in Proceedings of the 9th OpenFOAM Workshop, Zagreb, Croatia, 2014.
5. D. Destarac and A. Dumont, “ONERA M6 Wing Test-Case, Original and TMR”, NASA, 2022.
6. I. H. Abbott and A. E. von Doenhoff, “Theory of Wing Sections: Including a Summary of Airfoil Data”, Dover Publications, 1959.
7. 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,” 成功大學碩士論文, 2023.
8. O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation”, in Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, pp. 234–241, 2015.
9. R. Caruana, “Multitask Learning”, Machine Learning, Vol. 28, No. 1, pp. 41–75, 1997.
10. C. Lei, Y. Jiang, K. Xu, S. Liu, H. Cao and C. Wang, “Convolutional Neural Network Models for Visual Classification of Pressure Ulcer Stages: Cross-Sectional Study”, JMIR Medical Informatics, Vol. 13, p. e62774, 2025.
11. J. Zhang, Y. Xie, Q. Wu and Y. Xia, “Medical Image Classification Using Synergic Deep Learning”, Medical Image Analysis, Vol. 54, pp. 10–19, 2019.
12. A. M. Abdelmonem, M. M. Ata, A. A. Atey, A. A. Shaalan and R. A. EI-Sayed, “High-Performance Scene Classification in Remote Sensing Imagery Using a Custom Deep CNN Architecture”, Scientific Reports, Vol. 16, No. 1, p. 5946, 2026.
13. M. Yang and X. Fan, “YOLOv8-Lite: A Lightweight Object Detection Model for Real-Time Autonomous Driving Systems”, ICCK Transactions on Emerging Topics in Artificial Intelligence, Vol. 1, No. 1, pp. 1–16, 2024.
14. R. Kühlechner, “Object Detection Survey for Industrial Applications with Focus on Quality Control”, Production Engineering, Vol. 19, No. 6, pp. 1271–1291, 2025.
15. B. U. Mahmud, G. Y. Hong, A. A. Mamun, E. P. Ping and Q. Wu, “Deep Learning-Based Segmentation of 3D Volumetric Image and Microstructural Analysis”, Sensors, Vol. 23, No. 5, p. 2640, 2023.
16. X. Hui, J. Bai, H. Wang and Y. Zhang, “Fast Pressure Distribution Prediction of Airfoils Using Deep Learning”, Aerospace Science and Technology, Vol. 105, p. 105949, 2020.
17. H. Chen, L. He, W. Qian and S. Wang, “Multiple Aerodynamic Coefficient Prediction of Airfoils Using a Convolutional Neural Network”, Symmetry, Vol. 12, No. 4, p. 544, 2020.
18. Y. Zhang, W. J. Sung, and D. N. Mavris, “Application of Convolutional Neural Network to Predict Airfoil Lift Coefficient”, in Proceedings of 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2018.
19. Y. Y. Liu, J. X. Shen, P. P. Yang and X. W. Yang, “A CNN-PINN-DRL Driven Method for Shape Optimization of Airfoils”, Engineering Applications of Computational Fluid Mechanics, Vol. 19, No. 1, p. 2445144, 2025.
20. H. Moin, H. Z. I. Khan, S. Mobeen and J. Riaz, “Airfoil’s Aerodynamic Coefficients Prediction Using Artificial Neural Network”, in Proceedings of the 19th International Bhurban Conference on Applied Sciences and Technology (IBCAST), IEEE, pp. 123–128, 2022.
21. Y. Zhang, “A Better Autoencoder for Image: Convolutional Autoencoder”, in Proceedings of the International Conference on Neural Information Processing (ICONIP) Workshops. [Online]. Available: http://users.cecs.anu.edu.au/~Tom.Gedeon/conf/ABCs2018/paper/ABCs2018_paper_58.pdf (Accessed: Jun. 22, 2026), 2018.
22. N. Thuerey, K. Weibenow, 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.
23. Y. Lee, J. Park, J. Kim, J. H. Woo and J. H. Lee, “Conditional UNet Emulation of CMAQ Simulations for Fine Particulate Matter Concentration Prediction”, Scientific Reports, Vol. 15, No. 1, p. 38616, 2025.
24. A. Crespo Márquez, “The Curse of Dimensionality”, Digital Maintenance Management: Guiding Digital Transformation in Maintenance, Springer, pp. 67–86, 2022.
25. D. Ko, H. Ji, and Y. Ju, “Prediction of 3D Velocity Field of Reticulated Foams Using Deep Learning for Transport Analysis”, Transport in Porous Media, Vol. 148, No. 3, pp. 577–604, 2023.
26. J. Baxter, “A Model of Inductive Bias Learning”, Journal of Artificial Intelligence Research, Vol. 12, pp. 149–198, 2000.
27. J. Zhou, Q. Yu, C. Luo and J. Zhang, “Feature Decomposition for Reducing Negative Transfer: A Novel Multi-Task Learning Method for Recommender System (Student Abstract) ”, in Proceedings of the AAAI Conference on Artificial Intelligence, pp. 16542–16543, 2023.
28. A. Abolvardi, L. Hamey, and K. Ho-Shon, “UNET-Based Multi-Task Architecture for Brain Lesion Segmentation”, in Proceedings of the Digital Image Computing: Techniques and Applications (DICTA), IEEE, pp. 1–6, 2020.
29. C. Zhang, Z. Hu, Y. Shi and G. Xu, “Fast Aerodynamic Prediction of Airfoil with Trailing Edge Flap Based on Multi-Task Deep Learning”, Aerospace, Vol. 11, No. 5, p. 377, 2024.
30. S. Abuhanieh, B. Bicer, and M. Sahin, “A Validation for the OpenFOAM-Hisa Solver for Drag Prediction”, in Proceedings of the 32nd International Conference on Parallel Computational Fluid Dynamics, Nice, France, 2021.
31. F. R. Menter, “Improved Two-Equation k-omega Turbulence Models for Aerodynamic Flows”, NASA, TM-103975, 1992.
32. P. Cook, M. Mcdonald, and M. Firmin, “Experimental Data Base for Computer Program Assessment”, AGARD, Advisory Report No. 138, 1979.
33. A. Kuzmin, “On the Lambda-Shock Formation on ONERA M6 Wing”, International Journal of Applied Engineering Research, Vol. 9, No. 20, pp. 7029–7038, 2014.
34. N.V. Smirnov, “On the estimation of the discrepancy between empirical curves of distribution for two independent samples”, Bulletin de l'Université d'État de Moscou, vol. 2, no. 2, pp. 3–14, 1939.
35. 林哲甫, “主動式深度學習方法進行三維超音速流場預測,” 成功大學碩士論文,2024.
36. O. Malyshev, “Math at the Heart of CNN”, [Online]. Available: https://svitla.com/blog/math-at-the-heart-of-cnn/ (Accessed: Jun. 1, 2026), 2024.
37. M. Yani, S. B. Irawan, and S. C. Setiningsih, “Application of Transfer Learning Using Convolutional Neural Network Method for Early Detection of Terry’s Nail”, Journal of Physics: Conference Series, IOP Publishing, p. 12045, 2019.
38. P. Ramachandran, B. Zoph, and L. Q. Le, “Swish: A Self-Gated Activation Function”, arXiv preprint arXiv:1710.05941, 2017.
39. T. Gong, T. Lee, C. Stephenson, V. Renduchintala, S. Padhy and A. Ndirango, “A Comparison of Loss Weighting Strategies for Multi Task Learning in Deep Neural Networks”, IEEE Access, Vol. 7, pp. 141627–141632, 2019.
40. S. Liu, E. Johns, and A. J. Davison, “End-to-End Multi-Task Learning with Attention”, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11–20, 2019.
41. A. Kendall, Y. Gal, and R. Cipolla, “Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics”, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7482–7491, 2018.