Research On Fault Monitoring Technology Of Belt Conveyor Based On Distributed Fiber Optic Acoustic Sensing

Coal resources,as an important energy source,are irreplaceable for economic development and social progress.Belt conveyor is one of the important equipment for production and transportation in coal mines,and its normal operation is crucial for ensuring production and safety.However,due to the complex structure of belt conveyor and the harsh working environment,its failure rate is relatively high.Therefore,it is particularly important to monitor the faults of belt conveyor in coal mines.In the past,fault monitoring methods mostly relied on point sensors or switch nodes such as circuit breakers,which had blind spots in monitoring and were difficult to locate faults.Distributed acoustic sensing(DAS)technology using optical fiber as the sensing medium has the advantages of strong anti-electromagnetic interference ability,small size,inherent safety,and long-distance real-time monitoring,which can meet the requirements of working in harsh environments underground,and can be used for fault monitoring of belt conveyor.However,due to the strong noise interference and harsh application environment often present in coal mine sites,it is necessary to have effective signal processing and pattern recognition techniques to discriminate the fault signals.In this thesis,based on the actual coal transportation site of a coal mine in Datong,Shanxi Province,China,a fault monitoring scheme for belt conveyor based on 1DRes Net+SVM recognition algorithm is proposed.The deep mining of fault signal features is carried out to achieve effective identification of fault signals.The research content includes:(1)Starting from the Rayleigh scattering mechanism,the photoacoustic modulation mechanism is derived.A distributed optical fiber acoustic sensing system is built,and field data is collected;a fault and normal signal database for belt conveyor is constructed.(2)For the belt conveyor operation signal containing strong noise,the signal is preprocessed and feature analysis is performed.The signal is preprocessed using traditional wavelet denoising and high-pass filtering,and a total of 24-dimensional feature vectors are analyzed in time domain,frequency domain,and time-frequency domain.(3)A fault recognition algorithm based on 1DRes Net+SVM is proposed.By combining the residual structure with one-dimensional convolution,the input of the one-dimensional vibration signal collected by the DAS system is directly used without data conversion,effectively solving the problem of model degradation caused by deepening the network.A 1DRes Net structure that is superior to other classic one-dimensional convolutional neural networks is proposed.Three models are established: a 1DRes Net model based on high-pass filtering,a 1DRes Net model based on 24-dimensional time-frequency features,and a 1DRes Net model based on original signals.It is experimentally proved that the 1DRes Net model based on original signals has better performance.(4)Experiments are conducted to compare the performance of multiple machine learning classifiers such as decision trees,and the Softmax layer used for classification in the original network is replaced by a support vector machine(SVM)classifier with stronger classification performance.The experimental results show that the event recognition rate of the proposed 1DRes Net+SVM model based on original signals can reach 99.8%,which meets the requirements of practical applications.