摘要
模态参数广泛应用于拉索索力识别与状态评估,其自动化识别是拉索实时健康监测的关键技术之一。贝叶斯FFT算法和深度神经网络可实现拉索模态参数识别的自动化和不确定性分析,其中,贝叶斯FFT算法可快速识别拉索模态参数且定量其不确定性,但需人工输入结构自振频带信息。为此,基于Retinanet模型和一维卷积神经网络,设计提出一种轻量化频带选择网络,使之能自动检测拉索的振动频带,进而提出一种拉索模态参数全过程自动化识别方法。提出的频带选择网络模拟人工频带选择过程,融合多频率分辨率下功率谱曲线估计信息,通过检测功率谱曲线峰值确定拉索振动频带。为验证所提算法的有效性,利用沪苏通长江大桥拉索施工阶段实测加速度数据构建训练集和测试集,完成拉索模态参数的连续自动识别。试验证明,频带自动选择网络准确性超过95%,单样本平均推理时间达到毫秒级,与贝叶斯FFT相结合可实现拉索模态参数的自动化实时识别。另外,通过拉索频率的连续识别成功检测到拉索张拉事件,验证了基于频率对拉索损伤识别的潜在可能性。
Modal parameters have been widely used for force identification and condition assessment of cables.Automated modal parameter identification is one of the key requirements in the real-time health monitoring of cables.Although it is capable of identifying structural modal parameters and quantifying the identification uncertainty,Bayesian FFT algorithm requires an input of frequency bands to be initialized,which are determined manually based on professional knowledge,thus,limiting its automation.A method of combining Bayesian FFT algorithm with deep neural network is proposed for automatic identification of modal parameter and uncertainty quantitation.In this work,a frequency-band selection network is proposed based on the Retinanet and one-dimensional convolutional neural network.It imitates the manual frequency-band selection by taking advantage of estimated power spectrum density curves with different frequency resolutions.Furthermore,the proposed method is verified by measured data from Shanghai-Suzhou-Nantong Yangtze River Bridge.It is shown that the proposed method can achieve an accuracy higher than 95%with an average computation time of milli-seconds per sample in determining the frequency band.In addition,the cable tension adjustment was successfully detected through a continuous identification of cable frequencies,validating the potential usage of frequencies in damage detection of cables.
作者
安永辉
薛志林
廖子涵
李宾宾
An Yonghui;Xue Zhilin;Liao Zihan;Li Binbin(Dalian University of Technology,Dalian 116023,China;Zhejiang University,Haining 314400,China)
出处
《土木工程学报》
EI
CSCD
北大核心
2023年第7期82-91,共10页
China Civil Engineering Journal
基金
国家自然科学基金青年项目(51908494)
国家优秀青年科学基金(52122803)
辽宁省兴辽英才计划(XLYC1907060)。
关键词
拉索
模态识别
贝叶斯FFT
深度神经网络
cables
modal identification
Bayesian FFT algorithm
deep neural network
作者简介
安永辉(1986-),男,博士,教授。主要从事结构健康监测､结构振动控制和新型结构方面的研究;薛志林(1993-),男,博士研究生。主要从事结构健康监测方面的研究;廖子涵(1996-),男,博士研究生。主要从事结构可靠度方面的研究;通讯作者:李宾宾(1986-),男,博士,研究员。主要从事结构健康监测､贝叶斯方法和结构可靠度等方面的研究。E-mail:bbl@zju.edu.cn。
