卷积神经网络在风洞天平静态校准中的应用  

Application of convolutional neural network in static calibration of wind tunnel balance

作　　者：汪运鹏 聂少军 王粤 姜宗林[1,2] WANG Yunpeng;NIE Shaojun;WANG Yue;JIANG Zonglin(State Key Laboratory of High Temperature Gas Dynamics,Institute of Mechanics,Chinese Academy of Sciences,Beijing 100190,China;School of Engineering Science,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院力学研究所高温气体动力学国家重点实验室,北京100190 [2]中国科学院大学工程科学学院,北京100049

出　　处：《空气动力学学报》2023年第3期25-32,I0001,共9页Acta Aerodynamica Sinica

基　　金：国家自然科学基金(11672357,11727901)。

摘　　要：风洞天平是气动试验中用于测量作用在模型上的空气动力载荷(力与力矩)的大小、方向和作用点的装置,测量结果的精准度与天平的静态校准性能直接相关,天平的静态校准是通过校测设备建立天平测量信号与所受气动载荷的映射关系。由于多分量风洞天平的各个分量间存在相互干扰,并且通常二次干扰和组合干扰会出现非线性特性,采用线性拟合方法会产生一定的误差,使得风洞天平静态校准性能因受到数据处理方法(线性拟合)的局限而较难进一步提高。因此,为了进一步提升应变天平静态校准的性能,本文探索深度学习方法在风洞天平静态校准中的应用。利用中国科学院力学研究所风洞天平校准系统AiBCS,对六分量应变天平开展基于卷积神经网络的静态校准研究,采用深度学习训练模型代替传统风洞天平校准公式并获取更高性能指标。同时,对人工智能建模方法在天平静态校准中的适用条件、有效性及可靠性等方面进行了讨论和评估分析。数据结果显示:相较于传统的基于最小二乘多项式的拟合方法,卷积神经网络天平校准方法有效降低了天平各个分量间的载荷干扰,使校准结果的精准度得到了较大幅度的提升。As the most important measuring device in aerodynamic testing,the wind tunnel balance is used to measure the magnitude,direction,and point of the aerodynamic loads(forces and moments)acting on the test model.The accuracy of the measurement is directly related to the static calibration of the wind tunnel balance,which establishes the mapping relationship between the balance output signals and aerodynamic loads on the calibration equipment.This paper explores the possibility of improving the calibration performance of the strain-gauge balance in the calibration system AiBCS,developed by Institute of Mechanics of Chinese Academy of Science,using the convolutional neural network(CNN).The applicable conditions,validity,and reliability of CNN in the balance calibration are discussed and evaluated.Results obtained by the CNN-based calibration method and the traditional polynomial fitting method are analyzed and compared.It turns out that the CNN-based calibration method can effectively reduce the load interference between various balance components,yielding a significantly improved performance.Consequently,the deep-learning technology shows great application potential in calibrating wind tunnel balance.

关 键 词：气动测量 风洞 应变天平 静态校准 深度学习 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] V211.752[自动化与计算机技术—控制科学与工程]