Research On Fault Diagnosis Of Conveyor Belt Wear In Coal Preparation Plant Based On Machine Vision

Under long-term and high-load operation,the mining belt conveyor will produce obvious wear layers on the surface of the belt.If it is not monitored,once the wear is intensified,the belt will be worn through or broken,which will adversely affect production and bring economy Loss,or even endanger personal safety,causing serious production accidents.At present,there are many studies on belt safety protection,but most of them focus on acute failures such as belt tearing and deviation.The chronic failures mentioned above are not involved.The purpose of this study is to replace the daily manual inspection of belt wear and provide a more reliable,more effective,and smarter method for diagnosing belt wear faults,therefore,it has better novelty and practical value.Belt wear will form hierarchical wear layers.Most of the wear layered stripes are a few straight stripes parallel to the conveying direction of the belt conveyor.Wear layers range from 0 to 5.Take pictures of the belt wear and classify them.A total of 144 pieces Wear pictures.Due to the complex industrial site environment,some belt wear stripes are annular,and some worn belts will produce disruptive stripes on one or both sides of the belt.These conditions have caused greater interference in the detection of belt wear.First,the gray scale analysis method is proposed.After averaging the grayscale of the belt image parallel to the direction of the wear scar,the average grayscale curve is drawn with the abscissa is perpendicular to the direction of the wear scar,the ordinate is the average gray.Fitting the average gray curve to the sine function and calculating the residual of the average gray and the sine function can easily determine the number of belt wear layers to identify the degree of belt wear.However,because the belt wear stripes in the industrial field are not all linear,and there are still circular stripes,the grayscale analysis method has defects in judging the wear degree of such belts.Secondly,a diagnosis method based on artificial image feature extraction is proposed.Extract 7 texture features of belt wear pictures(4 grayscale co-occurrence matrix texture features and 3 Tamura texture features),2 grayscale color features and 2 image statistical features based on the characteristics of belt wear image(image entropy and image Standard deviation).A total of 11 image features,using a support vector machine(SVM)and linear discriminant analysis(LDA)to classify and learn the extracted image features according to the number of wear layers.Experiments show that the classification accuracy rate based on the SVM method is 83.8%,and the classification accuracy rate based on the LDA method is 82.4%.Finally,a diagnosis method based on deep learning is proposed.The 144 pictures on the industrial site are divided into 6 categories according to the number of wear layers.The migration training is performed with the AlexNet deep neural network and the GoogLeNet deep neural network.The ratio of the AlexNet data set to the test set is 8: 2.Besides,14 different wear images are used as detection sets.Since the 5-layer wear-out image data sets are few,the 5-layer wear-out data set is expanded by using four methods: Gaussian noise,fisheye transform,Fourier low-pass filtering and histogram equalization.GoogLeNet was used for migration training on the unexpanded data set and the expanded data set respectively.Experiments show that the average accuracy of AlexNet migration training is 91.5%,the average accuracy of GoogLeNet migration training is 87.7% for the unexpanded data set,and the average accuracy is 89.2% after expanding the 5-layer wear dataset.It can be seen that the AlexNet deep neural network method has the best recognition effect and can be used for the diagnosis of the chronic failure of belt wear.