摘要
结构仅输出的振动信号往往是各种源信号通过复杂规律形成的混合信号,对结构损伤特征提取与数据挖掘造成了很大困难。对此,提出了一种基于盲源分离(BSS)理论的复杂度追踪(CP)算法结合差分进化(DE)改进灰狼(GWO)算法优化的支持向量机(SVR)用于解决复杂结构的模态与损伤识别;CP算法基于信号预测性函数通过使分离信号的时间预测性度量最大化找到其线性混合矩阵,使分离分量具有最小复杂度并据此估计源信号。利用CP算法对结构响应信号进性分离得到信号分布向量(SDV)与分离源信号,通过定义差值曲率分布向量可以对结构损伤位置进行准确定位;对于损伤程度的识别,提出了一种DE改进的GWO对SVR进行优化的算法,即在GWO算法迭代过程中利用差分进化思想引入动态缩放因子以及交叉概率因子提高搜索和收敛速度,扩大种群所搜范围;利用不同工况下CP算法提取的差值曲率分布向量对结构损伤程度进行识别。通过对海上风电基础结构数值模型的分析,结果表明:CP算法对于高阶模态参数识别较fastICA表现出较强的适应性与优越性;同时,DE-GWO能够提高收敛速度,通过SVR算法对损伤的识别结果相比于BP神经网络更加准确。
The output-only vibration signals of structures are often mixed signals formed by various source signals through complex rules,which cause great difficulties in structural damage feature extraction and data mining.Therefore,a complexity pursuit(CP)algorithm based on the blind source separation(BSS)theory combined with the method of DE(differential evolution)-GWO(Grey Wolf optimization)-SVR(support vector regression)was proposed to solve the modal and damage identification of complex structures.The CP algorithm found the linear mixing matrix based on the signal predictability function by maximizing the temporal predictability measure of the separated signals,so that the separated components were of minimum complexity and the source signals were estimated accordingly.The CP algorithm was used to separate structural response signals into signal distribution vectors and separated source signals.By defining the difference curvature distribution vectors,the structural damage location was accurately determined.For the identification of damage extent,the GWO improved by DE was introduced to optimize SVR,that is,DE was used to improve the search and convergence speed and expand the search range of the population in the iterative process of the GWO algorithm.The difference curvature distribution vectors extracted by the CP algorithm under different conditions were used to identify the structural damage extent.The analysis on the numerical model of an offshore wind power infrastructure shows that the CP algorithm has stronger adaptability and superiority to higher-order modal parameter identification than the fastICA.At the same time,DE-GWO can improve the convergence speed,and the damage identification results by the SVR algorithm are more accurate than those by BP neural network.
作者
杜尊峰
邵玄玄
王晓梅
DU Zunfeng;SHAO Xuanxuan;WANG Xiaomei(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300354,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2020年第22期110-118,共9页
Journal of Vibration and Shock
基金
国家自然科学基金项目(51109158,51621092)
天津市自然科学基金(19JCYBJC21900)。
