The Research On Steel Cord Conveyor Belt Fault Detection Algorithm Based On Deep Learning

Steel cord conveyor belt is the traction and transport part of belt conveyor.As one of the most effective methods to detect the internal steel rope core faults of the coal mine conveyor belts,X-ray detection method uses the target detection algorithm to locate and identify the faults in the X-ray images,which not only achieves the realtime detection to the faults of the steel cord conveyor belts,but also reduces human resources costs.The thesis research object is focusing on the coal mine steel cord conveyor belts,based on deep learning a fault detection algorithm of steel cord conveyor belt.The following are the main contents studied in this dissertation:(1)Gathering fault images of steel cord conveyor belt and analyzing fault signature.After analyzing the structure of the detection system based on the principle of X-ray measurement,the ZSX-127 D mining X-ray detection apparatus is used to collect the images of many conveyor belts,the fault areas are marked in the images,the data set is made,the geometric characteristics and distribution pattern of faults in steel wire rope core conveyor belts are analyzed.(2)Establishing the FCOS fault detection model of steel core conveyor belt.The neural network structure and principles of the target detection algorithm are analyzed,the results show that the anchor-free detection algorithm can reduce the computational load and ensure detection precision,which can dramatically decrease the uneven distribution of positive and negative samples.And it is no longer limited by the fault scale and length-width ratio of the anchor frame to the wire rope core conveyor belt.Combined with FCOS network structure,the regression strategy,loss function and feature fusion network was analyzed.The FCOS algorithm has been determined as a deep learning fault detection model in this thesis.(3)Designing the steel cord conveyor belt fault detection improved algorithm based on FCOS.Combined with the geometric characteristics of the conveyor belt failures' marker boxes,the geometric factors such as the overlapping area of the bounding box,the aspect ratio,and the center point of the detection target are considered.In order to optimize the algorithm regression,the original loss function is replaced with CIo U in the improved algorithm.Adding feature fusion path in the feature fusion network to make full use of the characteristics of multi-scale feature fusion network,such as precise positioning and strong semantics,a new feature fusion network is proposed to optimize the poor performance of small targets which is easy to miss detection.The result of algorithm visualization has showed that the improved algorithm proposed has increased the precision by 20.9% on the basis of the original one,which has also met the demands of actual working conditions.Those advantages mentioned above has already verified the precision and feasibility of this algorithm.