Page 189 - 《振动工程学报》2026年第5期
P. 189
第 39 卷第 5 期 振 动 工 程 学 报 Vol. 39 No. 5
2026 年 5 月 Journal of Vibration Engineering May 2026
视 觉 振 动 测 量 的 稀 疏 传 感 方 法 研 究
王振宇 , 刘德庆 , 袁 杰 , 王 兴 1,3
2
1
1
(1. 中山大学航空航天学院,广东 深圳 518107; 2. 南安普顿大学计算工程设计研究组,英国 南安普敦 SO17 1BJ;
3. 深圳市智能微小卫星星座技术与应用重点实验室,广东 深圳 518107)
摘要:在复杂结构的高分辨率视觉振动测量中,需要采集与处理的视频数据量巨大,阻碍了该技术在航空航天结构振动试验
和结构健康监测中的应用。本文提出一种数据驱动的离线训练与在线重建密集测点振动数据结合的稀疏传感方法,在保证
振动测量结果精度的前提下,极大地减少了视觉振动测量中所需的测点数量。该方法通过采集高分辨率的结构振动视频,并
在结构表面设置密集测点,经光流分析得出密集测点振动数据,将其作为训练数据;通过本征正交分解实现数据降维,经正交
三角分解得到指定数量的稀疏测点位置,并建立密集测点与稀疏测点的振动响应映射关系;在实际结构振动测量过程中,仅
依靠少量选定的稀疏测点,较高精度地重建密集测点的振动响应。为验证所提出方法的有效性,本文开展了悬臂梁结构的视
觉振动测量,离线训练阶段采用随机激励的方式,减少了 76.2% 测点数量的同时保障重建训练数据精度为 8.43%。在线重建
阶段,采用随机激励、力锤激励、定频激励三种不同的工况考察所提方法的泛化能力,结果显示响应重建精度分别为 11.79%、
7.97% 和 7.08%,验证了该方法的数据压缩效率与振动测量精度。
关键词: 稀疏传感;数据重建;数据驱动;视觉振动测量;测点选取
中图分类号:O329;TP391.41 文献标志码:A DOI:10.16385/j.cnki.issn.1004-4523.202404017
Sparse sensing approach for visual vibrometry
2
1
1
WANG Zhenyu ,LIU Deqing ,YUAN Jie ,WANG Xing 1,3
(1.School of Aeronautics and Astronautics,Sun Yat-sen University,Shenzhen 518107,China;
2.Computational Engineering Design Group,University of Southampton,Southampton SO17 1BJ,UK;
3.Shenzhen Key Laboratory of Intelligent Microsatellite Constellation,Shenzhen 518107,China)
Abstract: In high-resolution visual vibrometry of complex structures, the vast amount of video data to be collected and processed has
hindered the practical application of this technology in structural vibration testing and health monitoring for aerospace structures. This paper
proposes a data-driven sparse sensing approach that combines offline training and online reconstructing the vibration data of dense points,
significantly reducing the number of required measuring points in visual vibrometry while ensuring the accuracy of the measurement results.
The approach first captures high-resolution structural vibration videos and sets up dense virtual measuring points on the structure surface,
obtaining vibration data of the dense points through optical flow analysis as training data. Subsequently, data dimensionality reduction is
achieved through proper orthogonal decomposition, and positions of sparse measuring points are obtained through orthogonal triangular
decomposition,establishing a mapping relationship of vibration responses between dense and sparse points. Finally,in the actual structure
vibration measurement process,only a small number of selected sparse measuring points are used to reconstruct the vibration responses of
dense measurement points. To verify the effectiveness of the proposed approach, this paper conducts visual vibration measurement of a
cantilever beam structure. In the offline phase,under random excitation,the proposed method reduces approximately 76.2% of the number of
measuring points while maintaining reconstruction accuracy around 8.43%. In the online phase,three different operating conditions—random
excitation,hammer excitation,and fixed-frequency excitation are considered,and the resulting online reconstruction errors are 11.79%,
7.97% and 7.08%,respectively. It highlights the efficiency of data reduction and the accuracy of response reconstruction of the proposed
approach.
Keywords:sparse sensing;data reconstruction;data-driven;visual vibrometry;measuring points selection
收稿日期:2024-04-07;修订日期:2024-08-26
基金项目:国家自然科学基金资助项目 (12072378);深圳市科技计划项目 (RCYX20210706092137055,ZDSYS20210623091808026,
202206193000001,20220815155101002)

