基于多任务学习的翼型结冰与气动特性预测

doi:10.7527/S1000-6893.2025.32983

Abstract

Abstract:

To address the problem of limited accuracy in single-task learning for airfoil icing prediction and the difficulty in simultane-ously obtaining icing morphology and aerodynamic characteristics， a multi-task learning based airfoil icing prediction framework called MTAF （Multi-Task Airfoil Former） is proposed. The framework employs an encoder-decoder architecture that extracts deep representations of airfoil geometry and flight conditions through a shared feature encoder， realizes feature fusion and task interaction via a task correlation module， and accomplishes icing prediction and aerodynamic coefficient prediction respectively through dedicated prediction heads. The performance of Transformer and ConvNeXt backbone net-works within this framework is comparatively studied. Based on training and testing with 625 NACA 2412 airfoil samples， results demonstrate that the Transformer architecture achieves pixel accuracy of 99.79% and coefficient of determination for 0.990 6 in icing prediction tasks， significantly outperforming 99.50% and 0.977 2 of ConvNeXt. For aerodynamic coefficient prediction， Transformer achieves coefficient of determination values of 0.975 4 for lift coefficient and 0.987 4 for drag coefficient， which are comparable to 0.973 3 and 0.990 1 of ConvNeXt， respectively， with each architecture showing distinct advantages. The MTAF framework maintains stable prediction performance across different icing types and various flight conditions， providing a high-precision prediction tool for airfoil icing engineering applications.

Key words: aircraft icing, image prediction, aerodynamic characteristics, Transformer, convolutional neural network

CLC Number:

V211.3

Huizhu XIN, Chen LI, Weimin SANG, Liang HOU. Airfoil icing and aerodynamic characteristics prediction based on multi-task learning[J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(16): 132983.

Figures/Tables 20

Fig.1

Fig.2

Table 1

Fig.3

Fig.4

Fig.5

Table 2

Fig.6

Fig.7

Table 3

Fig.8

Fig.9

Fig.10

Table 4

Table 5

Fig.11

Fig.12

Fig.13

Fig.14

Table 6

References 47

[1]	张强，高正红. 基于神经网络的翼型积冰预测［J］. 飞行力学， 2011， 29（2）： 6-9.
	ZHANG Q， GAO Z H. Prediction of ice accretions based on the neural net［J］. Flight Dynamics， 2011， 29（2）： 6-9 （in Chinese）.
[2]	REEHORST A， ADDY H， COLANTONIO R. Examination of icing induced loss of control and its mitigations［C］∥ AIAA Guidance， Navigation， and Control Conference. Reston： AIAA， 2010.
[3]	李哲，徐浩军，薛源，等. 结冰对飞机飞行安全的影响机理与防护研究［J］. 飞行力学， 2016， 34（4）： 10-14.
	LI Z， XU H J， XUE Y， et al. Research on flight safety effect mechanism and protection for aircraft icing［J］. Flight Dynamics， 2016， 34（4）： 10-14 （in Chinese）.
[4]	孙瑞山，吴迪. 机翼结霜和薄冰的危险性分析［J］. 航空计算技术， 2017， 47（4）： 10-13.
	SUN R S， WU D. Risk analysis of frost and ice on wing［J］. Aeronautical Computing Technique， 2017， 47（4）： 10-13 （in Chinese）.
[5]	何磊，钱炜祺，易贤，等. 基于转置卷积神经网络的翼型结冰冰形图像化预测方法［J］. 国防科技大学学报， 2021， 43（3）： 98-106.
	HE L， QIAN W Q， YI X， et al. Graphical prediction method of airfoil ice shape based on transposed convolution neural networks［J］. Journal of National University of Defense Technology， 2021， 43（3）： 98-106 （in Chinese）.
[6]	LYNCH F T， KHODADOUST A. Effects of ice accretions on aircraft aerodynamics［J］. Progress in Aerospace Sciences， 2001， 37（8）： 669-767.
[7]	BRAGG M. Aircraft aerodynamic effects due to large droplet ice accretions［C］∥ 34th Aerospace Sciences Meeting and Exhibit. Reston： AIAA， 1996.
[8]	易贤，朱国林，王开春，等. 翼型积冰的数值模拟［J］. 空气动力学学报， 2002， 20（4）： 428-433.
	YI X， ZHU G L， WANG K C， et al. Numerically simulating of ice accretion on airfoil［J］. Acta Aerodynamica Sinica， 2002， 20（4）： 428-433 （in Chinese）.
[9]	HINTON G E， SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks［J］. Science， 2006， 313（5786）： 504-507.
[10]	柴聪聪. 基于深度学习的翼型结冰及其气动特性预测［D］. 成都：电子科技大学， 2021： 23-41.
	CHAI C C. Prediction of airfoil ice accretion and aerodynamic characteristics based on deep learning［D］. Chengdu： University of Electronic Science and Technology of China， 2021： 23-41 （in Chinese）.
[11]	何磊，钱炜祺，董康生，等. 基于卷积神经网络的结冰翼型气动特性建模［J］. 航空学报， 2023， 44（5）： 126434.
	HE L， QIAN W Q， DONG K S， et al. Aerodynamic characteristics modeling of iced airfoil based on convolution neural networks［J］. Acta Aeronautica et Astronautica Sinica， 2023， 44（5）： 126434 （in Chinese）.
[12]	李小龙，洪冠新. 一种基于神经网络的机翼结冰冰型预测方法［C］∥ 中国航空学会飞行力学学术交流会论文集. 北京：国防工业出版社， 2006： 98-103.
	LI X L， HONG G X. A neural network-based prediction method for wing icing shape［C］∥ Proceedings of the Annual Conference of Academic Exchanges on Flight Mechanics and Flight Tests. Beijing： National Defense Industry Press， 2006： 98-103 （in Chinese）.
[13]	OGRETIM E， HUEBSCH W， SHINN A. Aircraft ice accretion prediction based on neural networks［J］. Journal of Aircraft， 2006， 43（1）： 233-240.
[14]	CHANG S N， LENG M Y， WU H W， et al. Aircraft ice accretion prediction using neural network and wavelet packet transform［J］. Aircraft Engineering and Aerospace Technology， 2016， 88（1）： 128-136.
[15]	YI X， WANG Q， CHAI C C， et al. Prediction model of aircraft icing based on deep neural network［J］. Transactions of Nanjing University of Aeronautics and Astronautics， 2021， 38（4）： 535-544.
[16]	STRIJHAK S， RYAZANOV D， KOSHELEV K， et al. Neural network prediction for ice shapes on airfoils using iceFoam simulations［J］. Aerospace， 2022， 9（2）： 96.
[17]	屈经国，王强，彭博，等. 基于多模态融合的任意对称翼型结冰预测方法［J］. 航空动力学报， 2024， 39（1）： 49-56.
	QU J G， WANG Q， PENG B， et al. Icing prediction method for arbitrary symmetric airfoil using multimodal fusion［J］. Journal of Aerospace Power， 2024， 39（1）： 49-56 （in Chinese）.
[18]	屈经国，彭博，易贤，等. 基于深度神经网络的任意翼型结冰预测方法［J］. 空气动力学学报， 2023， 41（7）： 48-55.
	QU J G， PENG B， YI X， et al. Icing prediction method for arbitrary airfoil using deep neural networks［J］. Acta Aerodynamica Sinica， 2023， 41（7）： 48-55 （in Chinese）.
[19]	柴聪聪，易贤，郭磊，等. 基于BP神经网络的冰形特征参数预测［J］. 实验流体力学， 2021， 35（3）： 16-21.
	CHAI C C， YI X， GUO L， et al. Prediction of ice shape characteristic parameters based on BP nerual network［J］. Journal of Experiments in Fluid Mechanics， 2021， 35（3）： 16-21 （in Chinese）.
[20]	YU D H， HAN Z R， ZHANG B， et al. A fused super-resolution network and a vision transformer for airfoil ice accretion image prediction［J］. Aerospace Science and Technology， 2024， 144： 108811.
[21]	YU D H， HAN Z R， ZHANG B， et al. A multi-autoencoder fusion network for fast image prediction of aircraft ice accretion［J］. Physics of Fluids， 2022， 34（7）： 076107.
[22]	YU J， DAI Y T， LIU X K， et al. Unleashing the power of multi-task learning： A comprehensive survey spanning traditional， deep， and pretrained foundation model eras［EB/OL］. arXiv preprint： 2404.18961， 2024.
[23]	CRAWSHAW M. Multi-task learning with deep neural networks： A survey［EB/OL］. arXiv preprint： 2009.09796， 2020.
[24]	VASWANI A， SHAZEER N， PARMAR N， et al. Attention is all you need［C］∥ Proceedings of the 31st International Conference on Neural Information Processing Systems. New York： ACM， 2017： 6000-6010.
[25]	LIU Z， MAO H Z， WU C Y， et al. A ConvNet for the 2020s［C］∥ 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Piscataway： IEEE Press， 2022： 11966-11976.
[26]	DOSOVITSKIY A， BEYER L， KOLESNIKOV A， et al. An image is worth 16x16 words： Transformers for image recognition at scale［C］∥ 9th International Conference on Learning Representations. Appleton： ICLR， 2021.
[27]	LIU Z， LIN Y T， CAO Y， et al. Swin Transformer： Hierarchical vision transformer using shifted windows［C］∥ 2021 IEEE/CVF International Conference on Computer Vision （ICCV）. Piscataway： IEEE Press， 2021： 9992-10002.
[28]	XIE E Z， WANG W H， YU Z D， et al. SegFormer： Simple and efficient design for semantic segmentation with transformers［C］∥ 35th Annual Conference on Neural Information Processing Systems （NeurIPS）. 2021： 12077-12090.
[29]	CHENG B W， SCHWING A G， KIRILLOV A. Per-pixel classification is not all you need for semantic segmentation［C］∥ Proceedings of the 35th International Conference on Neural Information Processing Systems. New York： ACM， 2021： 17864-17875.
[30]	JAIN J， LI J C， CHIU M， et al. OneFormer： One transformer to rule universal image segmentation［C］∥ 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Piscataway： IEEE Press， 2023： 2989-2998.
[31]	HENDRYCKS D， GIMPEL K. Gaussian error linear units （GELUs）［EB/OL］. arXiv preprint： 1606.08415， 2016.
[32]	XIONG R B， YANG Y C， HE D， et al. On layer normalization in the transformer architecture［C］∥ Proceedings of the 37th International Conference on Machine Learning. New York： ACM， 2020： 10524-10533.
[33]	HE K M， ZHANG X Y， REN S Q， et al. Deep residual learning for image recognition［C］∥ 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Piscataway： IEEE Press， 2016： 770-778.
[34]	MURPHY K， SCHÖLKOPF B， SRIVASTAVA N， et al. Dropout： A simple way to prevent neural networks from overfitting［C］∥ New York： ACM， 2014： 1929-1958.
[35]	HOWARD A G， ZHU M L， CHEN B， et al. MobileNets： Efficient convolutional neural networks for mobile vision applications［EB/OL］. arXiv preprint： 1704.04861， 2017.
[36]	BA J L， KIROS J R， HINTON G E. Layer normalization［EB/OL］. arXiv preprint： 1607.06450， 2016.
[37]	LARSSON G， MAIRE M， SHAKHNAROVICH G. FractalNet： Ultra-deep neural networks without residuals［C］∥ 5th International Conference on Learning Representations. Toulon： ICLR， 2017.
[38]	CARION N， MASSA F， SYNNAEVE G， et al. End-to-end object detection with transformers［C］∥ Computer Vision—ECCV 2020. Cham： Springer， 2020： 213-229.
[39]	ZEILER M D， FERGUS R. Visualizing and understanding convolutional networks［M］∥ Computer Vision—ECCV 2014. Cham： Springer， 2014： 818-833.
[40]	LIN T Y， GOYAL P， GIRSHICK R， et al. Focal loss for dense object detection［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2020， 42（2）： 318-327.
[41]	RASCHKA S， MIRJALILI V. Python machine learning： Machine learning and deep learning with Python， scikit-learn， and TensorFlow 2［M］. Birmingham： Packt Publishing Ltd， 2019： 20-23.
[42]	SHIN J， BOND T. Results of an icing test on a NACA 0012 airfoil in the NASA Lewis Icing Research Tunnel［C］∥ 30th Aerospace Sciences Meeting and Exhibit. Reston： AIAA， 1992.
[43]	BROEREN A P， WHALEN E A， BUSCH G T， et al. Aerodynamic simulation of runback ice accretion［J］. Journal of Aircraft， 2010， 47（3）： 924-939.
[44]	FAA. Part 25—Airworthiness standards： Transport category airplanes Appendix C： 4080-29-FR-18291［R］. Washington， D.C.： Federal Aviation Administration， 1970.
[45]	LOSHCHILOV I， HUTTER F. SGDR： Stochastic gradient descent with warm restarts［C］∥ 5th International Conference on Learning Representations. Toulon： ICLR， 2017.
[46]	LOSHCHILOV I， HUTTER F. Decoupled weight decay regularization［C］∥ 7th International Conference on Learning Representations. New Orleans： ICLR， 2019.
[47]	Aircraft Icing Technology Committee. Icing wind tunnel interfacility comparison tests： AIR5666A ［S］. Warrendale： SAE International， 2018.

参数	Case 1	Case 2
飞行速度/（m·s^-1）	67.05	67.05
环境温度/℃	-28.4	-6.1
液态水含量/（g·m^-3）	1.0	1.0
水滴平均直径/μm	20	20
结冰时长/s	360	360
飞行迎角/（°）	4	4

参数	数值
水滴平均直径/μm	20， 25， 30， 35， 40
液态水含量/（g·m^-3）	0.2， 0.4， 0.5， 0.6， 0.8
环境温度/K	261， 264， 265， 266， 268
飞行速度/（m·s^-1）	30， 40， 50， 60， 70
飞行迎角/（°）	2
结冰时长/s	1 080

参数	Transformer	ConvNeXt
结冰任务PA/%	99.84	99.62
结冰任务R²	0.992 4	0.982 1
升力系数R²	0.930 7	0.937 0
阻力系数R²	0.993 8	0.993 6
训练时长/s	349.72	324.11

参数	Transformer	ConvNeXt
像素准确率/%	99.82	99.58
R²	0.986 7	0.972 9

几何参数	测试集	ConvNeXt		Transformer
几何参数	测试集	RMSE	R²	RMSE	R²
冰形面积S_ice	62	8.22×10^-5	0.891 4	2.24×10^-5	0.991 9
冰形面积S_ice	100	5.05×10^-5	0.895 5	1.95×10^-5	0.984 5
驻点冰形高度h_sp	62	8.75×10^-4	0.981 4	5.55×10^-4	0.992 5
驻点冰形高度h_sp	100	8.18×10^-4	0.978 6	6.24×10^-4	0.987 6
上冰角高度h_u	62	8.70×10^-4	0.968 9	3.09×10^-4	0.996 1
上冰角高度h_u	100	1.23×10^-3	0.934 1	6.46×10^-4	0.981 9
下冰角高度h_l	62	8.90×10^-4	0.978 6	5.46×10^-4	0.992 0
下冰角高度h_l	100	8.42×10^-4	0.975 5	6.69×10^-4	0.984 5

Airfoil icing and aerodynamic characteristics prediction based on multi-task learning

RichHTML

PDF (PC)

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 20

References 47

Related Articles 15

Recommended Articles

Metrics

Comments

模型	C_L		C_D
模型	RMSE	R²	RMSE	R²
Transformer	0.004 1	0.958 1	0.001 3	0.970 1
ConvNext	0.003 9	0.961 6	0.001 2	0.973 4

[1]	Xiangyu WANG, Xianyun CHEN, Mengqi LIU, Zhenping FENG. Influence of vibration on aerothermal characteristics of turbine blade tips under maneuvering loads [J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(8): 132710-132710.
[2]	Jiayi LI, Haiqing SI, Hao DONG, Jingxuan QIU, Tianyu XIA. Verification and validation of deep dynamic stall numerical simulation method [J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(6): 132671-132671.
[3]	Tianhao GUO, Lei ZHANG, Xuefeng DUAN, Zhitao LIU, Fan LIU, Renyu ZHU. Aerodynamic and flight control characteristics evaluation of a flying boom based on wind tunnel tests [J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(16): 132934-132934.
[4]	Xiaobin SHEN, Zekun YE, Jingyu ZHAO, Jietao GUO, Guiping LIN. A three-dimensional surface water droplet impingement characterization method based on Eulerian streamlines [J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(11): 632751-632751.
[5]	Kaiqiang CAO, Pan CHENG, Xiangrong JING, Zhenbing LUO, Tianxiang GAO, Wenjie FENG, Yan ZHOU, Wenqiang PENG. Mechanism and experiment on de-icing by plasma synthetic jet actuator with oblique jet outlet [J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(11): 633381-633381.
[6]	Wangqing ZHU, Chenkai CAO, Guoqing ZHAO, Qinghua ZHU, Haoyu HU. Aerodynamic interference of scissor tail rotor and influence of unconventional layout parameters [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(S1): 732181-732181.
[7]	Lingling CHEN, Yang ZHANG, Yongqiang SHI, Qingzhen YANG. Film cooling performance/nozzle performance/infrared radiation characteristics of a vector nozzle [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(8): 631139-631139.
[8]	Lei WANG, Qiuxuan YE, Jin ZHANG. Spectrum sensing method for DME using combined wavelet transform and deep convolutional neural network [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(24): 332044-332044.
[9]	Dong GAO, Pujian LAI, Shilei WANG, Gong CHENG. RGB-T UAV object tracking based on feature-cooperative reconstruction [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(23): 632017-632017.
[10]	Jiaxin LI, Shuaishuai LYU, Yezi WANG, Yu YANG, Ziyue LI. Transformer-based intelligent tracking method of aviation structure surface cracks [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(21): 532355-532355.
[11]	Jiakun FAN, Junqiang AI, Ningjuan DONG, Jiakuan XU, Lei QIAO, Junqiang BAI. Stationary crossflow induced transition prediction method for supersonic swept-wing based on convolutional neural networks [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(20): 532012-532012.
[12]	Yinxuan ZHANG, Qi ZHANG, Zhenyong XU, Linshu MENG. Predicting method of aircraft mechanical response based on residual neural networks [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(19): 531295-531295.
[13]	Yao LU, Zihao LI, Zhunga LIU, Yanbo YANG. Intelligent data association and tracking of different types of multiple targets based on Transformer [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(17): 331643-331643.
[14]	He JIA, Wei JIANG, Wenlong BAO, Xin XU, Wei RONG, Li YU. Transonic/supersonic aerodynamic characteristics and fluid-structure interaction mechanism of flexible parachutes for planetary exploration [J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(1): 630369-630369.
[15]	Liping WANG, Fuxin WANG, Hong LIU. Research progress on simulation methods of drop diameter distribution in supercooled large drop icing conditions [J]. Acta Aeronautica et Astronautica Sinica, 2024, 45(S1): 730570-730570.