摘要
为适应结构健康监测对模态分析高效和自动化的需求,自主工作模态分析(Automated operational modal analysis,AOMA)在近二十年已成为结构动力学领域的研究热点之一。自动鉴别并剔除模态分析结果中的虚假模态,保留结构真实模态,是AOMA的核心和难点。现有AOMA方法可分为三类:(1)聚类法,核心是基于聚类算法自动挑选稳定图中的稳定模态作为真实模态,是目前的主流方法。(2)峰值提取法,自动挑选由真实模态引起的响应功率谱密度函数曲线或其奇异值曲线的峰值。(3)深度学习法,利用神经网络将工作模态分析问题转化为图像目标检测或时序分析问题。重点分析了三类方法的核心思想、分析流程及各自的特点,介绍了AOMA方法在工业软件及工程结构中的应用案例,最后阐明AOMA应用中应注意的问题及发展趋势。
Driven by the requirements of structural health monitoring for high efficiency and automation of modal analysis,the automated operational modal analysis(AOMA) has been the focused area in the field of structural dynamics over the past two decades.Automatically discriminating and eliminating the spurious modes while reserving real modes in the identified results plays a core role in the AOMA.The current AOMA methods are classified into three classes:(1)Clustering method.The clustering methods are used to automatically select the stable modes in the stabilization diagram as the real modes,which is the most widely-used method at present.(2)Peak picking method.It automatically picks the peaks,which are generated from the real modes,of power spectral density function or its singular values.(3)Deep learning method.The neural network is used to transform the operational modal analysis problem into object detection or time series analysis problem.The basic ideas,computational processes and the merits/demerits of three method classes are reviewed,and the applications of AOMA on commercial software and engineering structures are also introduced.Finally,the common issues and development trends of AOMA are presented.
作者
康杰
王寅
罗杰
孙嘉宝
曾舒洪
KANG Jie;WANG Yin;LUO Jie;SUN Jiabao;ZENG Shuhong(College of Astronautics,Nanjing University of Aeronautics and Astronautics,Nanjing 211106)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2023年第13期89-109,共21页
Journal of Mechanical Engineering
基金
国家自然科学基金青年资助项目(12102178)。
关键词
工作模态分析
自主辨识
虚假模态
聚类
峰值提取
深度学习
operational modal analysis
automated identification
spurious mode
clustering
peak picking
deep learning
