Research On Multi-dimensional Intelligent Inspection Method For Longitudinal Tearing Of Mining Conveyor Belt

Coal mine belt conveyor operating environment is bad,the conveyor belt as the most critical part of undertaking materials.The coal transported from the mining machinery to the place where the conveyor drops the material may be accompanied by sharp edges and corners(such as channel steel and Angle steel,etc.),and there is a height difference between the end and end of the two conveyor belts,sharp objects may directly penetrate or tear the conveyor belt,which will cause safety hazards.Longitudinal tearing mainly occurs at the loading point of the head and tail of the belt conveyor.It is mainly manifested as large cracks or complete tearing,which will reduce transportation efficiency and cause economic losses while producing potential transportation hazards.Therefore,it is of great significance to study the real-time detection of longitudinal tearing of conveyor belt.This paper firstly studies the detection technology of longitudinal tear of conveyor belt at home and abroad.From traditional machine learning to the latest conveyor belt surface defect recognition based on convolutional neural networks,CNN has been widely used in target detection tasks of mine operations with its powerful feature extraction ability and autonomous learning ability.Aiming at the shortcomings of current conveyor belt longitudinal tear target detection,such as single dimension,high model complexity and slow detection speed,a multi-dimensional real-time detection method of conveyor belt longitudinal tear based on deep learning network was proposed and implemented.Based on the experimental device of frequency conversion belt conveyor,longitudinal tear image data set was collected and made by industrial camera with multi-dimensional detection points.By improving the YOLOv4 target detection algorithm,Mobilenetv3 embedded in the mixed domain attention mechanism ECSNet was used as the backbone feature extraction network,and the lightweight integrated network model ECSMv3＿YOLOv4 was constructed.The multi-dimensional intelligent inspection prototype of belt conveyor was developed,the training,testing,identification and positioning experiments of network model were carried out.Based on the structural characteristics of ECSMv3＿YOLOv4 network model,the FPGA hardware accelerator architecture was customized.Winograd and GEMM fast computing engines were designed and the network recognition acceleration tests of PFGA platform were carried out for different light intensity sample images.By comparing the tests of CPU and different FPGA hardware accelerators,it is verified that the designed accelerators had obvious advantages in inference speed,power consumption and energy efficiency.The conveyor belt longitudinal tear intelligent monitoring system was developed to realize more efficient human-computer interaction tasks of damage monitoring and provide more intelligent inspection methods.Figure 55 Table 13 Reference 83...