| 引用本文: | 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.[点击复制] |
| HE Lei,QIAN Weiqi,YI Xian,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[点击复制] |
|
| |
|
|
| 本文已被：浏览 7527次 下载 5528次 |
| 基于转置卷积神经网络的翼型结冰冰形图像化预测方法 |
| 何磊1,钱炜祺1,易贤2,王强2,张显才1 |
|
(1. 中国空气动力研究与发展中心 计算空气动力研究所, 四川 绵阳 621000;2. 中国空气动力研究与发展中心 低速空气动力研究所, 四川 绵阳 621000)
|
| 摘要: |
| 结冰问题严重影响飞机飞行安全,结冰智能预测是飞机智能防除冰系统设计和安全设计的重要依据和支撑。为解决复杂冰形在翼面同一位置的法线方向冰形厚度存在多值的问题,提出基于转置卷积神经网络的翼型结冰冰形图像化预测方法。设计预测模型的神经网络结构、损失函数、数据规范等,直接将影响飞机结冰的飞行和大气条件作为输入,以灰度化的冰形图像作为输出。基于NACA0012翼型,通过数值模拟方法生成冰形数据集,同时利用风洞试验结果对数值模拟方法进行验证,以确保生成数据的可信度。构建以飞行速度、温度、液态水含量、平均水微滴直径和结冰时长5项参数作为输入的预测模型,并进行仿真训练和验证。仿真结果表明:所提翼型结冰预测模型不仅能够快速预测翼型冰形,而且在冰体轮廓、结冰上下极限、冰角位置、结冰厚度等主要特征方面也与数值计算结果符合较好。 |
| 关键词: 冰形 结冰 机器学习 深度学习 神经网络 翼型 预测 |
| DOI：10.11887/j.cn.202103013 |
| 投稿日期：2020-01-30 |
| 基金项目: |
|
| Graphical prediction method of airfoil ice shape based on transposed convolution neural networks |
| HE Lei1, QIAN Weiqi1, YI Xian2, WANG Qiang2, ZHANG Xiancai1 |
|
(1. Computational Aerodynamics Research Institute, China Aerodynamics Research and Development Center, Mianyang 621000, China;2. Low Speed Aerodynamics Research Institute, China Aerodynamics Research and Development Center, Mianyang 621000, China)
|
| Abstract: |
| Ice accretion affects flight safety severely. Intelligent ice accretion prediction is an important basis and support for the intelligent anti-icing and de-icing system design and safety design of aircraft. To solve the problem that there are multiple values of ice thickness in the normal direction of the same position of the airfoil surface with complex ice shape, a graphical ice accretion prediction method based on the transposed convolution neural network was proposed. The corresponding neural network structure, loss function and data specification of the prediction model were designed. The input of the prediction model was the data from flight and atmospheric conditions which directly affect ice accretion. The output of the prediction model is gray-scale image of ice shape. Ice shape data set was generated through numerical simulation method based on NACA0012 airfoil. To ensure the credibility of the generated data, wind tunnel test data was used to verify the numerical simulation method. Prediction model was established with five input parameters:liquid water content, median volumetric diameter, freezing time, temperature and flight speed, and was then trained and validated in simulation. The simulation results show that the proposed method can predict the ice shape quickly, and the main features of the predicted ice shape such as outline, upper and lower limits, position of ice horn and thickness fit well with the results of numerical simulation. |
| Keywords: ice shape ice accretion machine learning deep learning neural network airfoil prediction |
|
|
|
|
|
