Research On Structural Damage Feature Extraction Method Based On Robust Independent Component Analysis

With the development of the world economy and the increasing demand for technology,more and more civil engineering facilities are under construction and will be built in various countries,and the scale of the facilities is also increasing substantially.But at the same time,these civil structures are often faced with many security risks,such as long-term operation and use,unexpected natural disasters,bad environmental erosion,which not only cause the accumulation of structural damage,but also seriously damage the national economy and endanger people's lives and property.Therefore,the diagnosis and identification of structural damage has important theoretical and practical significance.Aiming at the problem of structural damage identification,this paper studies the method of structural damage feature extraction based on robust independent component analysis.The separation performance of kurtosis-based independent component analysis(kFastICA)algorithm and robust independent component analysis(RobustICA)algorithm is studied.On the basis of discussing the theory of independent component analysis,a large number of simulation experiments verify the separation characteristics of k-FastICA algorithm and RobustICA algorithm for super-Gaussian source signal,sub-Gaussian source signal and super-sub-gaussian mixture source signal respectively,and prove the high accuracy and fast convergence of RobustICA algorithm for separating super-Gaussian signal.At the same time,the effects of SNR and sampling points on the separation performance of the two algorithms are studied.Experiments show that the RobustICA algorithm still has high separation accuracy under the conditions of low SNR and few sampling points.Further experimental results show that RobustICA algorithm also shows a good separation effect for complex signals.RobustICA algorithm is used to extract structural damage features.Compared with kFastICA algorithm,RobustICA algorithm is superior to k-FastICA algorithm in structural damage feature extraction by using structural damage simulation data.The main manifestation is that RobustICA algorithm has higher independence and accuracy in separating components.At the same time,RobustICA algorithm is applied to the classical Passe II IASC-ASCE measured structural data.Through blind separation of 10 detection signals,the structure is not damaged under the excitation of environmental vibration and hammer impact,all inclined supports on the east side are removed,all inclined supports on the southeast side of all layers are removed,the inclined supports on the southeast side of the base and the third layer are removed,and the inclined supports on the southeast side of the base are removed.Except for the six working conditions of southeast inclined support and North inclined support of the first floor,the extraction accuracy is high.Furthermore,by calculating the stiffness ratio between damage condition and non-destructive condition,the accurate identification of damage degree under damage condition is realized.Aiming at the common problem of single channel blind source separation in practical engineering application,the feature extraction of single channel observation signal based on VMD-RobustICA algorithm is realized.Firstly,the variational mode decomposition(VMD)method is used to pre-process the complex single channel mixed signal,aiming at decomposing it into multiple IMF components.Then,the separated IMF components are used as the input of RobustICA algorithm,and then the mixed signals are separated to realize the feature extraction of the source signals.The method is applied to the measured bridge vibration data to extract the characteristic frequencies of the bridge cable forces under six conditions: no vehicle passing,a car passing,a big car passing,a continuous car passing,a continuous car passing,a mixed car passing.