基于EWT-FastICA的斜拉桥监测挠度温度效应分离

Separation of Cable-Stayed Bridge Monitoring Deflection Temperature Effect Based on EWT-FastICA

考虑桥梁挠度中的温度效应和长期挠度成分将会一定程度影响到桥梁的安全评估,提出基于经验小波变换(empirical wavelet transform,简称EWT)结合快速独立分量分析(fast independt component analysis,简称FastICA)方法对温度效应和长期挠度进行分离。首先,利用经验小波变换分离出日温差效应;其次,考虑年温差效应与长期挠度频率相近难以分离,因此运用经验小波变换自定间隔把傅里叶频谱上年温差和长期挠度部分划分成多个区间,并在每个区间内构造相应的小波滤波器,将单通道的挠度信号转化成无虚假模态的一系列本征模态函数(intrinsic mode function,简称IMF);然后,把多通道的IMF矩阵运用主成分分析(principal component analysis,简称PCA)降维;最后,将降维后的信号采用FastICA处理,实现桥梁挠度年温差和长期挠度的分离。数值仿真结果以及桥梁实测数据研究结果均表明:该方法能有效地分离挠度监测信号中的温度效应和长期挠度,且分离效率高。

Considering the temperature effect and long-term deflection components in bridge deflection will affect the safety assessment of bridges to a certain extent,a method based on empirical wavelet transform(EWT)combined with fast independent component analysis(FastICA)is proposed for the problem of temperature effect and long-term deflection separation.Firstly,the empirical wavelet transform is used to separate the effect of daily temperature difference;Secondly,considering that the frequency of the annual temperature difference is close to the frequency of the long-term deflection,it is difficult to separate,so the self-defined interval of empirical wavelet transform is used to divide the part of the annual temperature difference and long-term deflection in the Fourier spectrum into multiple intervals and construct corresponding wavelet filters in each interval.The deflection signal of a single channel is transformed into a series of intrinsic mode functions(IMF)without false modes;Then the principal component analysis(PCA)is applied to reduce the dimensionality of the multi-channel IMF matrix;Finally,the dimensionality-reduced signal is processed by FastICA to separate the annual tem perature difference and long-term deflection of the bridge deflection.The results of numerical simulation and bridge data survey prove that this method can effectively separate the temperature effect and long-term deflection in the monitoring deflection signal.