Study On Phase-sensitive Optical Time Domain Reflectometer And Event Classification Method

Phase-sensitive optical time domain reflectometer(?-OTDR) is one kind of novel distributed optical fiber sensing technology,which has the advantages of high sensitivity,high spatial resolution,long sensing distance and large dynamic range.Distributed acoustic sensors based on ?-OTDR have aroused increasing attention from both domestic and foreign researchers owing to its board range of industrial potential in the fields of cable state monitoring in the optical fiber transport network,pipeline security monitoring,border intrusion detection,structure health monitoring and micro-seismic wave detection.The optical fiber transport network itself doesn't have the ability of sensing the state of the transmission cable.That's why it's a difficult problem for network maintainers to constantly monitor and efficiently manage the network.We provide a practical ?-OTDR intelligent sensing network scheme for the above issue,which combines machine learning technology and distributed acoustic sensors based on ?-OTDR.The main research contents and contributions of this work are as follows:(1)A heterodyne coherent ?-OTDR data acquisition platform is designed and built.The conventional method is to directly sample the electrical signal acquired by heterodyne detection,and the sampling rate of the data acquisition is as high as 1 GS/s,which brings huge data amount.In this paper,the electrical signal is sampled after quadrature demodulation in the electrical domain,and the sampling rate can be reduced to 10 MS/s.So,data amount are largely reduced,meanwhile the cost of the system is decreased as well.The result of the external disturbance experiment shows that the disturbance locating method and the phase demodulation method can be implemented in the proposed platform with a sensing distance of ?10 km,a spatial resolution of ?10 m,and 1000 monitoring areas.(2)An ?-OTDR online data processing system is established.The phase change of the monitoring area with the external disturbance is obtained through real-time phase difference demodulation of 1000 monitoring areas including arctangent,phase difference,phase unwrapping,low frequency removal and channel selection by a field programmable gate array,and the correctness of the system is verified by integration test.The result of the audio sensing experiment shows that the platform with the data processing system has the ability of sound sensing.(3)An ?-OTDR offline event classification system is established.The data set of 5 sound events(using an air hammer,using an electric drill,using a metal hammer,using a wood saw,and using a pile driver)that come from construction sites contains 1000 pieces of the training data and 1000 pieces of the testing data.In this paper,ELM classifiers and SVM classifiers are introduced in the event classification.The classification performance,the noise immunity and the computing performance of the classifiers is compared.We point out that ELM classifiers promise to be an optimal choice in the ?-OTDR intelligent sensing network compared to SVM classifiers because of its high classification accuracy,less training time,less testing time and strong noise immunity.