Detection Of Crack On The Surface Of Bridge Based On Stacked Sparse Autoencoders And Distributed Strain

The detection of cracks on the bridge surface is of great significance to the safe operation of the bridge.The distributed optical fiber sensor can collect the strain distributed along the axial of the optical fiber,that is,the distributed strain.Distributed strain is sensitive to cracks,and the detection of cracks on the bridge surface can be achieved by detecting abnormal changes in the distributed strain.However,distributed strain often has problems such as low signal-tonoise ratio and difficulty in obtaining label data,which limits the application value of distributed strain in the detection of surface cracks on bridges.Deep learning is generally considered to improve the signal-to-noise ratio of data,and unsupervised classification methods can realize the detection of abnormal data.Therefore,it has certain theoretical and practical significance to carry out research on distributed strain-oriented deep learning noise reduction and unsupervised anomaly detection methods.This paper designs and implements two unsupervised crack detection methods based on deep learning.Firstly,inspired by the principle of image saliency detection based on contrast,an anomaly detection method based on stacked sparse autoencoder(SSAE)and global-local contrast fusion(GLC-Fusion)is proposed;the second is an anomaly detection method based on SSAE and cluster-based local outlier factor(CBLOF)crack detection method.Finally,experiments verify the effectiveness of the proposed method.In this article,the main work has the following three aspects:1.Distributed strain noise reduction methods,including Gaussian filtering,Robust Principal Component Analysis(RPCA)and SSAE.A SSAE model and method are proposed,which use SSAE to reconstruct the input distributed strain,which significantly improves the accuracy of the crack detection method.2.Research on unsupervised anomaly detection methods for distributed strains is carried out.The concept of image contrast is extended to the distributed strain data,and the unsupervised detection of cracks is realized by calculating the global and local contrast of the distributed strain data.Furthermore,in order to balance the contradiction between detection accuracy and calculation efficiency,an unsupervised crack detection method based on clusterbased local outlier factor(CBLOF)is explored.3.In bridge experiment.The distributed strain data on the surface of the bridge was collected on site,and the parameter comparison experiment involved in the method was systematically carried out.The results show that the proposed method can effectively detect the cracks on the bridge surface,and the method based on global-local contrast fusion(GLC-Fusion)is applicable for applications with higher accuracy requirements,the cluster-based local outlier factor(CBLOF)method is suitable for applications with higher real-time requirements.