飞机机体结构数字孪生技术通过高逼真度、动态更新的数字模型来反映飞机机体全寿命周期力学响应及综合性能。为了提高结构响应预测精度，数字孪生模型往往采用多层级、精细化仿真技术，同时也带来了模型计算效率和成本大幅增加的问题，难以满足飞行中对飞机结构强度性能实时预测的需求。基于人工智能的结构强度性能降阶预测是解决飞行中结构响应实时预测的关键技术手段，基于高阶数字孪生模型仿真结果，通过飞参、应变等实测数据信息推断的结构载荷，能够快速的对飞机结构力学响应进行高精度预测，可解决实际飞行中飞机结构性能预测的时效性问题，在飞机数字孪生技术领域已得到越来越多的重视。本文提出了一种将结构云图与载荷输入进行像素化处理的手段以表现点云数据间的空间关系，并结合引入了跨层连接机制的卷积神经网络，构造了一种以载荷为输入的力学响应智能预测方法。基于329种工况的机翼结构数值试验的结果表明：像素化处理方法能够在保留机体结构响应特征的同时使数据适配卷积等面向像素图的数据处理方式。采用的模型较传统卷积神经网络在预测损失与精度上均有提升，预测方法较传统仿真可实现近2个数量级以上的效率提升，预测应力分布与有限元仿真应力分布间相似度较高，具备在机体结构数字孪生中的应用价值。
Digital twin technology for aircraft structures reflects the mechanical response and comprehensive performance of an aircraft's lifecycle through high-fidelity, dynamically updated digital models. To improve the accuracy of structural re-sponse predictions, digital twin models often use multi-level, refined simulation techniques. However, this leads to a significant increase in model computation efficiency and cost, making it challenging to meet the real-time prediction needs for aircraft structural strength during flight. AI-based reduced-order prediction of structural strength is a key technology for real-time prediction of structural responses during flight. By combining high-order digital twin model simulation results with structural loads inferred from actual flight data such as flight parameters and strain, it is possi-ble to rapidly and accurately predict the mechanical response of aircraft structures. This addresses the timeliness is-sue of aircraft structural performance prediction during actual flights and has gained increasing attention in the field of aircraft digital twin technology. This paper proposes a method for pixelating structural cloud maps and load inputs to express spatial relationships between point cloud data. It also introduces a convolutional neural network with cross-layer connection mechanisms to construct an intelligent prediction method for mechanical responses based on load inputs. Results from numerical experiments on the wing structure under 329 conditions show that the pixelation meth-od can retain structural response characteristics while making the data compatible with pixel-based data processing methods like convolution. The model used shows improved prediction accuracy and reduced loss compared to tradi-tional convolutional neural networks. The prediction method achieves more than two orders of magnitude efficiency improvement compared to traditional simulations. The predicted stress distribution shows a high similarity to finite el-ement simulation stress distributions, highlighting its application value in the digital twin of aircraft structures.
[1] BURGESS N. Advances in numerical methods for CREATE-AV analysis tools[C]//52nd Aerospace Sci-ences Meeting. Maryland: AIAA, 2014: 0417.
[2] HALLISSY B P, HINE D, LAIOSA J P, et al. CREATE-AV quality assurance: best practices for vali-dating and supporting computation-based engineering software[C]//52nd Aerospace Sciences Meeting. Mary-land: AIAA, 2014: 0918.
[3] KRAFT E. HPCMP CREAT-AV and the air force digi-tal thread[C]//53rd AIAA Aerospace Sciences Meeting. Kissimmee: AIAA, 2015: 0042.
[4] FALLON T, MAHAL D, HEBDEN I. F-35 Joint strike fighter structural prognostics and health management: An overview[C]//Proceedings of the 25th symposium of the international committee on aeronautical fatigue. Rotterdam: The National Aerospace Laboratory (NLR), 2009: 1215.
[5] 张宝珍, 王萍. 飞机PHM技术发展近况及在F-35应用中遇到的问题及挑战[J]. 航空科学技术, 2020, 31(7): 18-26.
ZHANG B J, WANG P. Recent developments in aero-plane PHM technology and problems and challenges encountered in the application of F-35[J]. Aeronautical Science & Technology, 2020, 31(7): 18-26. (in Chinese)
[6] HEBDEN I G, CROWLEY A M, BLACK W. Over-view of the F-35 structural prognostics and health management system[C]//Proceedings of 9th European workshop on SHM. Manchester: The University of Manchester. 2018: 10-13.
[7] LIN M, GUO S, HE S, et al. Structure health monitor-ing of a composite wing based on flight load and strain data using deep learning method[J]. Composite Struc-tures, 2022, 286(8): 115305.
[8] SPILIOTOPOULOS P E, FERA F T, SARAMANTAS I E, et al. Development and experimental validation of a prototype system for Machine Learning based SHM in composite aerostructures[J]. Journal of Physics: Conference Series, 2024, 2692(1): 012025.
[9] JEONG S H, LEE K B, HAM J H, et al. Estimation of maximum strains and loads in aircraft landing using ar-tificial neural network[J]. International Journal of Aer-onautical and Space Sciences, 2020, 21(1): 117-132.
[10] MOHSEN R, KOUROSH H S, HAMED H K. Uncer-tainty quantification of aeroelastic wings flutter using an optimized machine learning approach[J]. Proceed-ings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, 2022, 236(15): 3167-3185.
[11] 常辉,朱靖,安朝,等.应用前馈神经网络的大柔性机翼阵风响应分析[J]. 计算机测量与控制, 2022, 32(8): 236-244.
CHANG H, ZHU J, AN C, et al. Analysis of gust re-sponse of large flexible wing using feedforward neural network[J]. Computer Measurement & Control, 2022, 32(8): 236-244. (in Chinese)
[12] HUANG W, WANG R, ZHANG M, et al. The research on deep learning-driven dimensionality reduction and strain prediction techniques based on flight parameter data[J]. Applied Sciences, 2024, 14(9): 3938.
[13] ZHANG Y, CAO S, WANG B, et al. A flight parame-ter-based aircraft structural load monitoring method us-ing a genetic algorithm enhanced extreme learning ma-chine[J]. Applied Sciences, 2023, 13(6): 4018.
[14] LI H, ZHANG Q, CHEN X. Deep learning-based sur-rogate model for flight load analysis[J]. Computer Modeling in Engineering & Sciences, 2021, 128(2): 605-621.
[15] 赵燕, 宋江涛, 唐宁. 某机翼的安全预测载荷模型建立[J]. 航空学报, 2020, 41(10): 259-266.
ZHAO Y, SONG J T, TANG N. Establishment of safety prediction load model for a wing[J]. Acta Aeronautica et Astronautica Sinica, 2020, 41(10): 259-266. (in Chi-nese)
[16] 金鑫, 殷建业, 王健志. 基于深度学习的飞行载荷测试与反演方法研究[J]. 航空工程进展, 2020,11(06): 887-893.
JIN X, YIN J Y, WANG J Z. Research on flight load test and inversion method based on depth learning[J]. Advances in Aeronautical Science and Engineering, 2020, 41(10): 259-266. (in Chinese)
[17] ZHANG W, DOI K, GIGER M L, et al. An improved shift‐invariant artificial neural network for computer-ized detection of clustered microcalcifications in digi-tal mammograms[J]. Medical Physics, 1996, 23(4): 595-601.
[18] GOODFELLOW I, BENGIO Y, COURVILLE A. Deep learning[M]. MIT Press, 2016: 413-417.
[19] HE K, ZHANG X, REN S, et al. Deep residual learn-ing for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: Computer Vision Foundation, 2016: 770-778.
[20] IOFFE S, SZEGEDY C. Batch normalization: acceler-ating deep network training by reducing internal co-variate shift[C]//International Conference on Machine Learning. Lille: International Machine Learning Socie-ty, 2015: 448-456.
[21] PASCANU R, MIKOLOV T, BENGIO Y. On the diffi-culty of training recurrent neural net-works[C]//International Conference on Machine Learn-ing. Atlanta: International Machine Learning Society, 2013: 1310-1318.
[22] QI C R, SU H, MO K, et al. Pointnet: deep learning on point sets for 3d classification and segmenta-tion[C]//Proceedings of the IEEE conference on com-puter vision and pattern recognition. Honolulu: IEEE Computer Society, 2017: 652-660.
CJA