Study On Detection And Location System Of Conveyor Belt Surface Defect Based On Visual Analysis

Conveyor belts play a vital role in various scenarios of transportation.However,defects of conveyor belt can easily cause serious economic losses and major safety accident.At present,most of the industry are training workers to conduct regular inspections,but these methods are time-consuming,inefficient and expensive.Therefore,taking the conveyor belt in the complex environment of the coal mine industry as the object,this thesis studies how to use machine vision to replace the human eye to detect surface defects of the conveyor belt in real time,and locates them through a combination of speed sensor and optical flow method,which is convenient for workers to make fixed maintenance.The main works of this thesis are as follows:(1)Since there is no database on surface defects of conveyor belts at home and abroad,in order to verify the performance of the defect detection network,this thesis proposes a novel image synthesis method: based on a small number of existing defect samples,the idea of DCGAN(Deep Convolutional Generative Adversarial Networks,DCGAN)is adopted to generate a large number of diverse defects negative images.And the synthetic method is also used to establish a database of more than fifty thousand images,which has great application value for the future research on surface defects detection of conveyor belts.(2)The thesis analysis the characteristics and principles of Faster RCNN and YOLO that are widely used in target detection.After comparative experiments,it is determined that YOLOv2 is used as the surface defect detection algorithm.The experiments also prove that the accuracy of the detection network reached 95.2% in test data after training with the established database.Also,the K-means clustering method is introduced to determine the anchor scheme and the exponential linear unit activation function is adopted in feature extraction network,making it that the detection and location accuracy are improved by 0.5% and 1.1% respectively.Also,training time is saved by 16.7%.(3)For the part of locating defects,the thesis combines the hall type of contact sensor with the dense optical flow method and applies the accumulated distance of the conveyor belt to calculate the relative displacement so can get the defects specific position.This method is not only not affected by the run direction,but also limits the defect positioning error within ±10m.Besides,the simulation experiments show that the accuracy of locating reaches 96%,which meets the location requirements of the long-distance conveyor belt for maintenance.(4)A complete system for detecting surface defects is proposed and a friendly interactive interface is also designed in this thesis.The system can quickly and accurately detect surface defects and the man-machine interface can clearly display the location of them,which also includes functions such as alarm classification and historical report generation.The spot tests are taken to verify the feasibility and validity of our research achievement,which of high practical value.In summary,based on computer vision and sensor technology,we study the detection and location surface defects of conveyor belt in complex environments.It provides a new idea for practical application in industry.47 Figures,8 Tables and 70 References.