火箭推进2025，Vol.51Issue(6)：67-76,10.DOI:10.3969/j.issn.1672-9374.2025.06.007
基于学习的视频运动放大算法在振动测试中的应用
Applications of learning-based video motion magnification algorithm for vibration test
摘要
Abstract
In engine ground tests,vibration test of pipes and other key components is crucial for optimizing structural design and improving safety.To address the difficulty of vision-based methods to accurately capture vibration information in micro-amplitude vibration tests,this paper first introduces the learning-based video motion magnification(LB-VMM)to magnify the pixel changes in micro-amplitude vibration videos.Then,the Kanade-Lucas-Tomasi(KLT)optical flow theory is adopted to track the motion so that the magnified vibration signals can be extracted for time or frequency domain analysis.Finally,two practical application paths,vibration signal analysis and rapid mode shape identification,are further constructed.Vibration test of the pipeline and mode shape identification test of the beam are carried out for verification.Compared with the phase-based video motion magnification,the LB-VMM magnified video significantly reduces motion artifacts and distortions,and the computation speed is improved by about 20 times.By introducing LB-VMM to magnify the micro-amplitude vibration,gains can be made to the high-frequency components.The micro-amplitude vibration video can directly visualize the high-quality mode shape after the time-domain band-pass filtering and high magnification with LB-VMM,and the MAC values between the experimental mode shapes and theoretical mode shapes of the beam are above 0.9.The findings demonstrate that LB-VMM can effectively improve the ability of vision-based methods for testing the micro-amplitude vibration,with a good application prospect.关键词
火箭发动机/计算机视觉/微幅振动/视频运动放大/深度学习/光流Key words
rocket engine/computer vision/micro-amplitude vibration/video motion magnification/deep learning/optical flow分类
航空航天引用本文复制引用
张一鸣,郝方楠,徐自力,李广,金旭..基于学习的视频运动放大算法在振动测试中的应用[J].火箭推进,2025,51(6):67-76,10.基金项目
装备重大基础研究项目(514010106-302) （514010106-302）
