Research On Belt Conveyor Fault Diagnosis System Based On Sound Signal

Belt conveyor has become an irreplaceable transport tool in port coal transport industry because of its advantages of large transport volume,long transport distance and continuous transportation.With the gradual increase of the demand for coal,the belt conveyor will work for a long time with high load.Once the equipment fails,the economic losses will be immeasurable.At present,fault diagnosis based on belt conveyors mainly relies on manual inspection methods,which are no longer suitable for the current production status due to its large workload and low diagnostic efficiency.Through field research,the sound signal which are generated by the broken belt conveyor contains a lot of fault information.In this thesis,a set of fault diagnosis system for belt conveyor based on sound signal is developed by analyzing the sound produced in the broken belt conveyor,The main content of this paper is as follows:(1)Aiming at the noise interference of belt conveyor,an improved wavelet threshold denoising method is proposed.After selecting the wavelet basis,decomposition layer number and threshold value of sound signal,and analyzing the convergence of soft and hard threshold function,I propose a wavelet threshold denoising method based on improved threshold function,and make a comparative experiment with traditional wavelet modulus maximum denoising method.The simulation results show that the improved method can filter out the noise more effectively and reduce the distortion of the sound signal.(2)Aiming at the problem that traditional sound characteristics cannot better characterize signal characteristics,two features,Mel-Frequency Cepstral Coefficients(MFCC)and Deep Learning,were extracted from the sound signals.The extraction of MFCC is based on the fast Fourier transform of the denoised sound signal,which is filtered by Mel filter and obtained by logarithmic transformation.Feature extraction based on Deep Learning is to transform the sound signal into spectrogram by short-time Fourier transform,and then put it into the neural network for extraction.The information which can best represent the signal type in the sound signal of the belt conveyor is obtained through the extraction of two kinds of features.(3)In order to solve low accuracy of fault diagnosis model of belt conveyor based on convolutional neural network and support vector machine,a deep learning model of VGG16 based on convolutional neural network is built firstly,and the structure of VGG16 is improved according to the sample data.The VGG16 neural network with the best recognition ability is obtained through the optimization design of learning rate and batch sample size by putting the sample data of sound signal visualization into the VGG16 neural network.Secondly,the fault diagnosis model based on SVM is built,which respectively takes the improvement of feature-level fusion and decision-level fusion as control experiments.Feature-level fusion is the use of principal component analysis to reduce the dimensions of the two extracted features and then merge them.Decision-level fusion is the use of D-S evidence theory to fuse the posterior probabilities obtained by the two classifiers.Through the fault diagnosis comparative experiments of three SVM methods,it is concluded that the recognition rate of SVM model based on decision-level fusion is higher,which is also better than the improved VGG16 neural network.