An intelligent belt wear fault diagnosis method based on deep learning

Belt conveyors are important transportation equipment in coal mining enterprises. At present, most research on this topic focuses on areas such as tear resistance and foreign body identification. Few studies have focused on belt wear, but belt wear is the subject of daily inspections on site. The artificial grayscale analysis method, support vector machine (SVM) method, and deep learning network are proposed herein to identify the degree of belt wear by using image acquisition devices installed on belt conveyors to collect images of no-load belts, instead of manual inspection. The experimental results indicate that the grayscale analysis method has limitations in identifying belt wear. For complex types of wear, such as annular wear, its recognition capability is poor, and the grayscale analysis method is heavily dependent on the results of human analysis. The highest accuracy of the SVM method is 84.5%, and it effectively identifies complex wear states. After training, worn belts can be detected automatically. However, the selection of features during training completely depends on human decisions, and the accuracy is affected by such factors that have a great influence. The deep learning network attained a 91.5% average recognition accuracy rate with the highest accuracy being 95%. It can fully automate intelligent feature selection, training and detection.