Research On Key Technologies Of Foreign Object Detection System For Belt Conveyor

As the key equipment for coal transportation,belt conveyors often encounter noncoal foreign objects due to the complex production environment and improper human operation,which can easily scratch the high-speed running belt,causing safety accidents and economic losses.At present,in some large mines,although a relatively complete centralized monitoring system has been established in the belt conveyor transportation working face,personnel still need to detect foreign objects through underground monitoring videos,which has a low level of automation and can easily lead to missed and incorrect detection of foreign objects;At the same time,the coal transportation line is too long,and a large amount of redundant information in the video is transmitted and stored,increasing the work cost of cloud servers.Therefore,in order to improve the intelligent level of coal transportation,this article takes the monitoring images of belt conveyors as the detection basis,and aims to achieve the detection of non-coal foreign objects on the conveyor belt.Foreign object detection is carried out on the coal transportation images on the belt conveyor,determining the category and location of foreign objects,and terminal detection of foreign objects on the belt conveyor is achieved through embedded devices.The main work and research content of the thesis are as follows:(1)Design of the overall framework of the foreign object detection system for belt conveyors.Analyzed the underground coal transportation process and main equipment,studied the system requirements and performance requirements for foreign object detection of belt conveyors based on the characteristics of underground monitoring video images,built the overall architecture of the belt conveyor foreign object detection system,proposed a frame preprocessing method for underground monitoring video based on frame difference method,and sorted out the foreign object detection process of belt conveyors.(2)Research on an improved YOLOX based foreign object detection algorithm for belt conveyors.Analyzed the basic principle of YOLO series object detection algorithms,collected and established a foreign object image dataset for belt conveyors,enhanced image sample data,introduced IAT image enhancement module and CBAM attention mechanism,designed a rotation decoupling head,predicted the angle information of foreign objects,and constructed a MO-YOLOX network structure;The foreign object detection model was trained on the foreign object image dataset of the belt conveyor,and related experiments were conducted to test the performance of MOYOLOX foreign object detection,verifying the effectiveness of the foreign object detection algorithm for the belt conveyor.(3)Research on pruning and deployment methods for sparse foreign object detection models.Based on Grad-CAM,a visualization analysis of the foreign object detection model for belt conveyors was conducted.A pruning algorithm based on BN layer weight dynamic sparsity was proposed to prune the foreign object detection model.The pruned foreign object detection model was deployed to embedded devices,and comparative experiments were conducted on the pre pruning and post pruning foreign object detection models,as well as experiments on foreign object terminal detection under embedded devices.(4)Experimental research on foreign object detection system for belt conveyors.We have developed a foreign object detection system software for belt conveyors,and conducted ground and industrial experiments at the Provincial Collaborative Innovation Center for Intelligent Mining Equipment and a certain coal mine in Henan,then tested the performance of the developed belt conveyor foreign object detection system on the embedded device Jetson Xavier NX and verified the applicability of the designed foreign object detection system to the working environment.In this thesis,there are 84 figures,25 tables and 90 references.