视觉跟踪多转子位移测量  

Vision tracking multi-rotor displacement measurement

作　　者：杨荣良 王森[1] 伍星 柳小勤[1] 刘韬[1] YANG Rong-liang;WANG Sen;WU Xing;LIU Xiao-qin;LIU Tao(School of Mechanical and Electrical Engineering,Kunming University of Science and Technology,Kunming 650500,China;Yunnan Mechanical and Electrical Vocational and Technical College,Kunming 650023,China)

机构地区：[1]昆明理工大学机电工程学院,云南昆明650500 [2]云南机电职业技术学院,云南昆明650023

出　　处：《振动工程学报》2024年第1期113-125,共13页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(52065035);云南省重大科技专项计划项目(202102AC080002)。

摘　　要：针对当前传统振动传感器在测量旋转体位移时受限于安装和测点数量等问题,将高速工业相机作为采集媒介,在转子振动试验台上进行转子振动视频的采集,并利用基于多目标跟踪的视觉振动测量方法跟踪多个转子目标的全场振动位移。将注意力机制引入残差神经网络,结合特征金字塔网络结构建立改进的特征提取骨干网络,并利用身份重新识别方法来强化相邻帧间目标位移的关联性,跟踪旋转体全场振动位移信号。在转子振动位移测量数据集上对不同网络模型进行定性和定量的比较。结果表明,本文构建的网络模型在边界框回归时能够获取更为紧密的贴合度;将采集的电涡流位移信号作为标准量进行两个转子位移信号的对比实验,结果表明,本文多目标跟踪算法拟合的波形和频谱噪声最小,且能与电涡流信号相匹配;在目标对象模糊情况下的实验也证明本文算法所具有的泛化性能,这也体现出视觉测量在旋转体振动位移跟踪领域的工程应用价值。Aiming at the problem that the current traditional vibration sensor is limited by the installation and the number of measur-ing points when measuring the displacement of the rotating body,this paper uses a high-speed industrial camera as the acquisition medium,and collects the rotor vibration video on the rotor vibration test bench.The visual vibration measurement method tracks the full-field vibration displacement of multiple rotor targets.The feature pyramid network structure is introduced into the residual neural network,and the improved feature extraction backbone network is established by combining the attention mechanism.The identity re-identification method is used to strengthen the correlation of target displacement between adjacent frames,and to track the full-field vibration displacement signals of the rotator.Qualitative and quantitative comparisons of different network models on the rotor vibration displacement measurement dataset show that the network model proposed in this paper can obtain a tighter fit when the bounding box is regressed.The collected eddy current displacement signal is used as the standard value to compare the two rotor displacement signals,and the experimental results show that the waveform and spectral noise fitted by the multi-target tracking algorithm in this paper is the smallest and can match the eddy current signal.The experiments also prove the generalization performance of the algorithm in this paper,which reflects the engineering application value of visual measurement in the field of vi-bration displacement tracking of rotating bodies.

关 键 词：视觉测振 深度学习 多目标 视觉跟踪 模糊图像 旋转体位移测量 

分 类 号：TH113.21[机械工程—机械设计及理论] TH74[自动化与计算机技术—计算机应用技术] TP391.41[自动化与计算机技术—计算机科学与技术]