基于CEEMD和统计参数的斜拉桥损伤识别方法研究

Damage identification method of cable-stayed bridges based on CEEMD and statistical parameters

为解决仅使用互补集成经验模态分解(complementary ensemble empirical mode decomposition,CEEMD)方法的斜拉桥信号分解存在含噪固有模态函数(intrinsic mode function,IMF)分量且不能进行损伤定量的问题,提出了一种基于CEEMD与统计参数方法相结合的斜拉桥损伤识别方法。该方法基于CEEMD方法对斜拉桥动力响应信号进行自适应性分解,确定适用的白噪声幅值标准差并推导CEEMD方法的集成次数,得到各阶IMF分量;采用欧氏距离对分解的IMF分量进行谱系聚类分析以避免模态混叠现象;采用峰度统计参数的有效权重峰度指标方法滤除含噪IMF分量,提取有效IMF分量并重构为有效IMF分量和;利用变异系数统计参数、二阶中心差分法和泰勒展开式推导损伤定位指标,根据四阶统计矩峰度统计参数推导损伤定量指标。用所提方法对某斜拉桥进行损伤识别研究,结果表明:仿真分析的损伤定位识别精度为100%,损伤定量最大误差为1.80%;在高斯白噪声干扰下,损伤定位不受影响,损伤定量最大误差为1.88%;进行实桥的损伤识别,结果表明实桥主梁无损伤。

Here,to solve problems of intrinsic mode function(IMF)components with noise and unable to quantify damage existing in cable-stayed bridges’signal decomposition obtained only with complementary ensemble empirical mode decomposition(CEEMD),a damage identification method for cable-stayed bridge based on combining CEEMD and statistical parameters method was proposed.This method could adaptively decompose dynamic response signals of cable-stayed bridge based on CEEMD,determine applicable amplitude standard deviation of white noise,and derive the integration times of CEEMD to obtain various orders of IMF components.Euclidean distance was used to perform spectral clustering analysis for the decomposed IMF components and avoid modal aliasing phenomena.The effective weight kurtosis index method for kurtosis statistical parameters was used to filter out noisy IMF components,extract effective IMF components,and reconstruct them for forming the sum of effective IMF components.Coefficient of variation statistical parameters,second-order central difference method and Taylor expansion were used to derive damage localization indexes.Damage quantification indexes were derived according to the 4-order statistical moment kurtosis statistical parameters.The proposed method was used to study damage identification of a cable-stayed bridge,and the results showed that the accuracy of damage localization and identification in simulation analysis is 100%;the maximum error in damage quantification is 1.80%;under Gaussian white noise interference,damage localization is not affected,and the maximum error in damage quantification is 1.88%;conducting damage identification of an actual bridge,no damage to its main beam is found.