摘要
针对平面电极传感器在薄冰探测过程中存在硬件设计复杂、厚度计算滞后和无法重复探测的问题,设计了一种可用于小型飞机的机翼结冰实时探测系统。选取具有高导热特点的氧化铝作为平面电极传感器基板材料,使用电容频率扫描法对不同厚度下的薄冰特性进行测量,并通过嵌入NUC平台中的神经网络模型完成对冰厚的实时预测。实验结果表明,在0~2 mm的探测厚度范围内,所建立模型的训练集、验证集和测试集在[-0.2 mm,+0.2 mm]误差区间内的预测概率分别为97.7%、96.3%和70.5%。为实现系统的重复性探测,进一步研究了不同初始温度下传感器表面的薄冰相变行为,为机翼覆冰厚度的高精度检测与实时测量提供了一种潜在的技术。
To address the problems of complex hardware design,lagging thickness calculation and inability to repeat detection in the process of thin ice detection by planar electrode sensors,a real-time wing ice detection system that can be used for small aircraft was designed.Aluminum oxide with high thermal conductivity was selected as the substrate material for the planar electrode sensor,and the capacitive frequency scanning method was used to measure the characteristics of thin ice at different thicknesses and to predict the ice thickness in real time using a neural network model embedded in the NUC platform.The experimental results show that the prediction probability of the model is 97.7%,96.3%and 70.5%for the training set,validation set and test set,respectively,within the error interval[-0.2 mm,+0.2 mm]for the detection thickness range of 0~2 mm.To achieve reproducible detection of the system,the phase change behaviour of thin ice on the sensor surface at different initial temperatures was further investigated,providing a potential technical means for high accuracy detection and real-time measurement of wing ice cover thickness.
作者
李泽军
杜振宇
李子寅
辛雍
张琳
邓霄
LI Zejun;DU Zhenyu;LI Ziyin;XIN Yong;ZHANG Lin;DENG Xiao(College of Optoelectronics Engineering,Taiyuan University of Technology,Taiyuan 030024,China;AVIC Taiyuan Aero-Instruments Co.,Ltd,Taiyuan 030006,China;College of Architecture,Taiyuan University of Technology,Taiyuan 030024,China;College of Physics,Taiyuan University of Technology,Taiyuan 030024,China)
出处
《仪表技术与传感器》
CSCD
北大核心
2023年第8期61-68,共8页
Instrument Technique and Sensor
基金
山西省应用基础研究计划(201901D111061)
山西省重点研发计划资助项目(201903D321001)。
关键词
机翼结冰
平面电极
频率扫描
神经网络
电热除冰
wing icing
planar electrodes
frequency scanning
neural network
electrothermal de-icing