-
摘要:
随着机器学习的快速发展和其突出的非线性映射能力，越来越多的学者将机器学习方法应用到流体力学领域。为克服传统数学拟合不能很好的解决系统非线性问题，以及现有文献中所提及的一些基于神经网络的气动参数预测方法，需要进行参数化处理而带来的不便，同时为实现多变量多输出气动参数快速预测的目的，基于卷积神经网络考虑机翼变迎角和浮沉建立了一种多变量多输出的机翼气动参数预测模型，实现了机翼气动参数的快速预测。结果表明：所建模型具有较高且稳定的预测精度，并且计算效率较计算流体力学(CFD)提高了40倍。
Abstract:With the rapid development of machine learning and its outstanding nonlinear mapping ability, more and more scholars apply machine learning methods to the field of fluid mechanics. To overcome the obstacle that the traditional mathematical fitting cannot well present the system nonlinearity and the inconvenience of some neural network-based aerodynamic parameter prediction methods due to the need of parametric processing, and to achieve the multi-variable and multi-output aerodynamic parameters, this paper establishes a multi-variable and multi-output model based on convolutional neural network considering the variable angle of attack and the heave of the wing to realize the rapid prediction of the aerodynamic coefficient of the wing. The results show that this model has high prediction accuracy and its computational efficiency is 40 times higher than computational fluid dynamics (CFD). Moreover, the designed stability experiment results show that the proposed model has good stability.
-
图 2 基于RBF网格变形机翼状态图及升力和力矩系数
Figure 2. Deformed wing state diagram and lift drag coefficient based on RBF mesh
表 1 差积神经网络具体结构和参数设置
Table 1. Specific structure and parameter setting of CNN
层名称 参数 输入层 机翼状态改变后的原始图像（200×200×3） 卷积层1 卷积核（5×5×96）+relu 池化层1 滤波器大小（3×3）strides（步长）=2 卷积层2 卷积核（3×3×96）+relu 池化层2 滤波器大小（3×3）strides（步长）=2 卷积层3 卷积核（3×3×128）+relu 池化层3 滤波器大小（3×3）strides（步长）=2 全连接层（并列2个） relu+dropout 输出层（并列2个） 激活函数tanh 
下载: 导出CSV
-
[1] 安效民, 徐敏, 陈士橹. 多场耦合求解非线性气动弹性的研究综述[J]. 力学进展, 2009, 39(3): 284-298. doi: 10.3321/j.issn:1000-0992.2009.03.004AN X M, XU M, CHEN S L. An overview of CFD/CSD coupled solution for nonlinear aeroelasticity[J]. Advances in Mechanics, 2009, 39(3): 284-298(in Chinese). doi: 10.3321/j.issn:1000-0992.2009.03.004 [2] 李亚东, 张子军, 张钧尧, 等. 电动飞机气动焦点辨识及飞行试验研究[J]. 航空工程进展, 2021, 12(3): 78-84. doi: 10.16615/j.cnki.1674-8190.2021.03.10LI Y D, ZHANG Z J, ZHANG J Y, et al. Research on aerodynamic center identification and flight test of electric aircraft[J]. Advances in Aeronautical Science and Engineering, 2021, 12(3): 78-84(in Chinese). doi: 10.16615/j.cnki.1674-8190.2021.03.10 [3] 原智杰, 张公平, 崔茅, 等. 基于神经网络的导弹气动参数预测[J]. 航空兵器, 2020, 27(5): 28-32. doi: 10.12132/ISSN.1673-5048.2019.0164YUAN Z J, ZHANG G P, CUI M, et al. Prediction of missile’s aerodynamic parameters based on neural network[J]. Aero Weaponry, 2020, 27(5): 28-32(in Chinese). doi: 10.12132/ISSN.1673-5048.2019.0164 [4] SURESH S, OMKAR S N, MANI V, et al. Lift coefficient prediction at high angle of attack using recurrent neural network[J]. Aerospace Science and Technology, 2003, 7(8): 595-602. doi: 10.1016/S1270-9638(03)00053-1 [5] CARPENTER M, HARTFIELD R, BURKHALTER J. A comprehensive approach to cataloging missile aerodynamic performance using surrogate modeling techniques and statistical learning[C]// 29th AIAA Applied Aerodynamics Conference. Reston: AIAA, 2011: 3029. [6] 刘昕. 基于神经网络的机翼气动参数预测仿真研究[J]. 计算机仿真, 2015, 32(12): 67-71. doi: 10.3969/j.issn.1006-9348.2015.12.014LIU X. Simulation of airfoil plunging aerodynamic parameter prediction based on neural network[J]. Computer Simulation, 2015, 32(12): 67-71(in Chinese). doi: 10.3969/j.issn.1006-9348.2015.12.014 [7] BALLA K, RUBEN S, OUBAY H, et al. An application of neural networks to the prediction of aerodynamic coefficients of aerofoils and wings[J]. Applied Mathematical Modelling, 2021, 96: 456-479. doi: 10.1016/j.apm.2021.03.019 [8] 叶舒然, 张珍, 王一伟, 等. 基于卷积神经网络的深度学习流场特征识别及应用进展[J]. 航空学报, 2021, 42(4): 524736.YE S R, ZHANG Z, WANG Y W, et al. Progress in deep convolutional neural network based flow field recognition and its applications[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 524736(in Chinese). [9] 陈海, 钱炜祺, 何磊. 基于深度学习的翼型气动系数预测[J]. 空气动力学学报, 2018, 36(2): 294-299.CHEN H, QIAN W Q, HE L. Aerodynamic coefficient prediction of airfoils based on deep learning[J]. Acta Aerodynamica Sinica, 2018, 36(2): 294-299(in Chinese). [10] 吴正文. 卷积神经网络在图像分类中的应用研究[D]. 成都: 电子科技大学, 2015: 8-12.WU Z W. Application research of convolution neural network in image classification[D]. Chengdu: University of Electronic Science and Technology of China, 2015 : 8-12 (in Chinese). [11] 李宏伟, 吴庆祥. 智能传感器中神经网络激活函数的实现方案[J]. 传感器与微系统, 2014, 33(1): 46-48. doi: 10.3969/j.issn.1000-9787.2014.01.012LI H W, WU Q X. Implementation scheme for activated function of neutral networks in intelligent sensors[J]. Transducer and Microsystem Technologies, 2014, 33(1): 46-48(in Chinese). doi: 10.3969/j.issn.1000-9787.2014.01.012 [12] 尹宝才, 王文通, 王立春. 深度学习研究综述[J]. 北京工业大学学报, 2015, 41(1): 48-59. doi: 10.11936/bjutxb2014100026YIN B C, WANG W T, WANG L C. Review of deep learning[J]. Journal of Beijing University of Technology, 2015, 41(1): 48-59(in Chinese). doi: 10.11936/bjutxb2014100026 [13] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386 [14] NAIR V, HINTON G E. Rectified linear units improve restricted boltzmann machines[C]//Proceedings of the 27th International Conference on Machine Learning, 2010: 807- 814. [15] 聂雪媛, 郑冠男, 杨国伟. 含间隙非线性机翼跨声速颤振时滞反馈控制[J]. 北京航空航天大学学报, 2021, 47(10): 1980-1988. doi: 10.13700/j.bh.1001-5965.2020.0356NIE X Y, ZHENG G N, YANG G W. Time delay feedback control for transonic flutter of airfoil with free-play nonlinearity[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(10): 1980-1988(in Chinese). doi: 10.13700/j.bh.1001-5965.2020.0356 -
下载:
点击查看大图
计量
- 文章访问数: 481
- HTML全文浏览量: 82
- PDF下载量: 64
- 被引次数: 0
