Research And Development Of On-line Detection System For Longitudinal Tear Of Conveyor Belt

Belt conveyor is an indispensable tool in the process of coal transportation.Longitudinal tear of conveyor belt will lead to major safety accidents.Therefore,the detection of longitudinal tear of conveyor belt is the key content of monitoring and management in coal mine safety production.At present,the main methods of longitudinal tear detection of conveyor belt are visual method based on edge segmentation to extract crack features and visual method based on threshold segmentation to extract light fringe features.Because of the harsh production environment in coal mine,these methods are usually not reliable.To solve this problem,this paper studies two methods of tear defect detection based on machine learning,and develops the corresponding real-time online detection system.Specific research contents are as follows:(1)A tear defect detection method based on SVM is proposed.Firstly,the real-time image is segmented and preprocessed,and then the candidate area of tearing is determined by sliding window technology.Finally,HOG features of candidate regions are used as input of SVM classifier to judge whether there is tearing in candidate regions.A large number of experiments show that the detection accuracy of this method is 86.2%,and the detection time of single frame image is about 26 ms.(2)A tear defect detection method based on CNN is proposed.Using a large number of data collected in the field,the main CNN-based target detection models such as Fater RCNN,Mask RCNN,YOLOv3 and SSD are trained,and the performance of each model is tested comprehensively.The test results show that YOLOv3 model has the best comprehensive performance compared with other models.Its single frame detection time is about 33 ms and detection accuracy is 89.2%.It basically meets the application requirements.(3)A real-time online detection system for longitudinal tear of conveyor belt is developed.The specific functions of the system include real-time image acquisition,single frame image tear detection,dynamic analysis of tear grade,etc.In addition,the system can upload relevant information to the host computer,or send out alarms and other control signals.Experiments show that the system has high reliability and strong practicability.