Distributed Strain-based Microcrack Detection Approach By Robust Principal Component Analysis And Support Vector Machine

Microcrack detections are of great significance to the health operation of bridges.The distributed optical fiber sensor based on Brillouin Optical Time Domain Analysis technology can measure distributed strains along length of the fiber with high spatial resolution and sensitivity.Since the distributed strain are sensitive to the cracks,it provides a powerful solution to the detection problem.However,due to the low signal-to-noise ratio(SNR)of the measurements,the strain corresponding to the cracks are prone to be confused with noise.In order to solve this confusion problem and realize the accurate microcrack detections,it has theoretical and practical significance to conduct research on distributed strain denoising and detection methods.A microcrack detection approach is proposed based on robust principal component analysis(RPCA)and support vector machine(SVM).The distributed strain are assumed consisting of two classes of strain sequences,one is measured from cracked structural elements and the other from crack-less structural elements.Therefore,the detection problem is considered as a binary class classification problem on the sequences.Firstly,the measured distributed strain are denoised by RPCA.Secondly,the distributed strains are segmented into a set of subsequences,which are with the same length.Finally,a SVM classifier with Gaussian kernel is trained and tested on the subsequences.The experimental results demonstrated that in laboratory the microcracks with crack opening displacement down to23 ?m can be detected,in field test on a bridge the cracks can also be detected with an accuracy up to 99.98%.It is concluded that the proposed approach can solve the confusion problem and has competitive noise robustness performance over traditional methods.The main work includes:1.The denoise method based on RPCA is proposed focusing on the low SNR issue of the measured distributed strain.A data matrix construction method is designed on the distributed strain.By exploiting RPCA,the data matrix is decomposed into a low-rank structure matrix and a sparse noise matrix.Quality of the distributed stain is improved by removing the noise matrix from the data matrix.2.The microcrack detection problem is transformed into a binary class classification problem,further solved by a Gaussian kernel SVM classifier.It is superior to multilayer perceptron(MLP)and decision tree on detection performance.3.The proposed approach is verified on in-lab beam tests and in-field bridge tests.