Abstract: Aerodynamic modeling of 3D aircraft under varying flight conditions is essential for aircraft optimization design. This study formulates the task as condition-guided regression of the surface pressure field using a point cloud-based deep surrogate model, in which surface mesh points, geometric attributes, and flight-condition parameters are jointly represented. The dataset covers Mach number from 3.0 to 10.0 and angle of attack from -10° to 10°. Within this framework, the key challenge is how to fuse condition and geometry features and how to aggregate local geometry while preserving generalization across configurations and data scales. To address this issue, we construct task-specific datasets, design representative static and dynamic fusion modules, and perform systematic ablation experiments on geometric aggregation backbones. The obtained results lead to two practical guidelines. First, condition information should be introduced through progressive static fusion, whereas late concatenation and attention-based reweighting require caution; modulation inspired by deep operator networks is more promising when sufficient data are available. Second, a hybrid aggregation backbone that uses dynamic aggregation in dense point cloud stages and static aggregation in sparse stages offers a better balance between accuracy and computational cost.

面向多工况场景的泛化飞行器压力场建模特征工程研究

[1]BhatnagarS, AfsharY, PanSW, et al., 2019. Prediction of aerodynamic flow fields using convolutional neural networks. Computational Mechanics, 64(2):525-545.

[2]ChenFG, AkasakaK, 2021. 3D flow field estimation around a vehicle using convolutional neural networks. The 32nd British Machine Vision Conference, article 396.

[3]ChenLW, CakalBA, HuXY, et al., 2021. Numerical investigation of minimum drag profiles in laminar flow using deep learning surrogates. Journal of Fluid Mechanics, 919:A34.

[4]DingF, LiuJ, ShenCB, et al., 2015. Novel inlet–airframe integration methodology for hypersonic waverider vehicles. Acta Astronautica, 111:178-197.

[5]DuruC, AlemdarH, BaranÖU, 2021. CNNFOIL: convolutional encoder decoder modeling for pressure fields around airfoils. Neural Computing and Applications, 33(12):6835-6849.

[6]EricksonLL, 1990. Panel Methods—an Introduction. NASA Technical Paper 2995, National Aeronautics and Space Administration, Washington, USA.

[7]EshaghiMS, AnitescuC, ThombreM, et al., 2025. Variational physics-informed neural operator (VINO) for solving partial differential equations. Computer Methods in Applied Mechanics and Engineering, 437:117785.

[8]GuoYL, WangHY, HuQY, et al., 2021. Deep learning for 3D point clouds: a survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(12):4338-4364.

[9]HinesD, BekemeyerP, 2023. Graph neural networks for the prediction of aircraft surface pressure distributions. Aerospace Science and Technology, 137:108268.

[10]KashefiA, RempeD, GuibasLJ, 2021. A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries. Physics of Fluids, 33(2):027104.

[11]KilicD, NikbayM, 2025. Implementation and assessment of machine learning methods for multidisciplinary aerospace problems. AIAA SCITECH 2025 Forum, article 0492.

[12]LiJC, DuXS, MartinsJRRA, 2022. Machine learning in aerodynamic shape optimization. Progress in Aerospace Sciences, 134:100849.

[13]LiuF, HanZH, ZhangY, et al., 2019. Surrogate-based aerodynamic shape optimization of hypersonic flows considering transonic performance. Aerospace Science and Technology, 93:105345.

[14]LiuJQ, ChenRQ, LouJH, et al., 2023. Deep-learning-based aerodynamic shape optimization of rotor airfoils to suppress dynamic stall. Aerospace Science and Technology, 133:108089.

[15]LuL, JinPZ, PangGF, et al., 2021. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators. Nature Machine Intelligence, 3(3):218-229.

[16]MaLQ, KongLY, PengXS, et al., 2024. PSTNet: transformer for aggregating neighborhood features in 3D point cloud semantic segmentation of eggplant plants. Scientia Horticulturae, 331:113158.

[17]MartinsJRRA, LambeAB, 2013. Multidisciplinary design optimization: a survey of architectures. AIAA Journal, 51(9):2049-2075.

[18]MassegurD, da RonchA, 2024. Graph convolutional multi-mesh autoencoder for steady transonic aircraft aerodynamics. Machine Learning: Science and Technology, 5(2):025006.

[19]MukherjeeNK, RapolS, UnnithanS, et al., 2020. Optimization of a 3D combustion bowl geometry using response surface modeling. SAE International Journal of Engines, 14(2):173-198.

[20]ParkT, LiuMY, WangTC, et al., 2019. Semantic image synthesis with spatially-adaptive normalization. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p.2332-2341.

[21]PerezE, StrubF, de VriesH, et al., 2018. FiLM: visual reasoning with a general conditioning layer. Proceedings of the AAAI Conference on Artificial Intelligence, p.3942-3951.

[22]QiCR, HaoS, MoKC, et al., 2017. PointNet: deep learning on point sets for 3D classification and segmentation. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), p.77-85.

[23]QianGC, LiYC, PengHW, et al., 2022. PointNeXt: revisiting PointNet++ with improved training and scaling strategies. Advances in Neural Information Processing Systems 35, p.23192-23204.

[24]SamaniegoE, AnitescuC, GoswamiS, et al., 2020. An energy approach to the solution of partial differential equations in computational mechanics via machine learning: concepts, implementation and applications. Computer Methods in Applied Mechanics and Engineering, 362:112790.

[25]ShaoXQ, LiuZJ, ZhangSQ, et al., 2023. PIGNN-CFD: a physics-informed graph neural network for rapid predicting urban wind field defined on unstructured mesh. Building and Environment, 232:110056.

[26]ShenY, HuangW, WangZG, et al., 2023. A deep learning framework for aerodynamic pressure prediction on general three-dimensional configurations. Physics of Fluids, 35(10):107111.

[27]ShenY, ZhangH, HuangW, et al., 2024. Geometric-perspective transfer learning for fast aerodynamic prediction in few-shot tasks. Physical Review Fluids, 9(10):104101.

[28]ShiSS, GuoCX, JiangL, et al., 2020. PV-RCNN: point-voxel feature set abstraction for 3D object detection. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p.10526-10535.

[29]ShihB, PeyvanA, ZhangZQ, et al., 2025. Transformers as neural operators for solutions of differential equations with finite regularity. Computer Methods in Applied Mechanics and Engineering, 434:117560.

[30]ShuklaK, OommenV, PeyvanA, et al., 2024. Deep neural operators as accurate surrogates for shape optimization. Engineering Applications of Artificial Intelligence, 129:107615.

[31]SunG, WangSY, 2019. A review of the artificial neural network surrogate modeling in aerodynamic design. Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, 233(16):5863-5872.

[32]VaswaniA, ShazeerN, ParmarN, et al., 2017. Attention is all you need. 31st Conference on Neural Information Processing Systems (NIPS 2017), p.6000-6010.

[33]XiongFF, ZhangL, HuX, et al., 2023. A point cloud deep neural network metamodel method for aerodynamic prediction. Chinese Journal of Aeronautics, 36(4):92-103.

[34]YuanZX, SongX, BaiL, et al., 2022. Temporal-channel transformer for 3D lidar-based video object detection for autonomous driving. IEEE Transactions on Circuits and Systems for Video Technology, 32(4):2068-2078.

[35]ZhangH, ShenY, HuangW, et al., 2025. Deep transfer learning for three-dimensional aerodynamic pressure prediction under data scarcity. Theoretical and Applied Mechanics Letters, 15(2):100571.

[36]ZhangY, HanZH, ShiLX, et al., 2016. Multi-round surrogate-based optimization for benchmark aerodynamic design problems. 54th AIAA Aerospace Sciences Meeting, article 1545.

[37]ZhaoHS, JiangL, JiaJY, et al., 2021. Point transformer. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), p.16239-16248.

Similar articles