Research On The Block Debris Detection System For Rail Grinding Vehicle Operation Based On Deep Learning

With the improvement of high-speed railway operation speed and transportation capacity,steel rails may suffer from damage and defects such as crushing,side wear,wave wear,peeling,and fat edges,posing hidden dangers to driving safety.Rail grinding,as one of the important means in line maintenance and repair,can slow down and prevent the occurrence of various diseases,thereby achieving the goals of extending the service life of the rail,improving the smoothness of vehicle operation,and reducing noise.The high-temperature grinding debris generated by the rail grinding vehicle during operation can form blocks.Currently,it mainly relies on manual equipment for processing,which has problems such as low efficiency,low intelligence,and high safety hazards.Therefore,in order to effectively solve the safety hazards caused by the flying high-temperature block shaped debris during grinding operations and the falling agglomerated debris during self operation,this article designs a block shaped debris detection system.Through the processing host and efficient algorithms for real-time online dynamic analysis,it achieves high-speed data collection and processing of the falling high-temperature block shaped debris,and timely notifies line maintenance personnel of the accurate mileage position results of the debris through wireless communication,To provide a basis for the rapid,fixed point,and accurate collection of high-temperature abrasive debris.The main work of this article is as follows:(1)Analysis of detection and inspection methods for block debris.Based on the working characteristics of the rail grinding vehicle in both operation and running conditions,the types,causes,distribution,and physical characteristics of grinding debris were analyzed.The advantages and disadvantages of existing technologies were compared in detail,and a deep learning based object detection algorithm was ultimately selected to detect block shaped debris.(2)Propose an improved YOLOv5 block debris detection algorithm.Firstly,through the visualization experiment of the middle layer feature map,the problem of insufficient effective feature extraction in the entire training process of the input image was analyzed.Secondly,by adding CBAM attention mechanism to the original network,this article further enhances the feature expression of block shaped debris;The Bi FPN structure replaces the original PAN structure of YOLOv5,enhancing the utilization of shallow features;Adding a large-scale detection layer can more accurately locate small-sized block debris;Select EIo U as the loss function of the target box regression to accelerate the convergence speed.Then,a self-made block debris dataset was used for training.Finally,a ablation experiment was conducted,and the YOLOv5 algorithm before and after improvement was compared through the visualization of intermediate layer feature maps and thermal analysis.The original algorithm,improved algorithm,and current mainstream object detection algorithms were compared and tested to demonstrate the superiority of the improved YOLOv5 block debris algorithm.(3)This paper proposes an optimization method based on DeepSort for block debris detection results,which addresses the problem of repeated statistics caused by the occurrence of block debris in multiple consecutive frames of images during the detection process.Firstly,the feature extraction network of the target tracking algorithm is retrained using the established dataset for block shaped debris re recognition;Secondly,the detection results of the improved YOLOv5 algorithm are fed into the DeepSort target tracking algorithm,and Kalman filtering is used to predict and update the target tracking box in real time.The data association between the previous frame tracking results and the current frame detection results is established through cascade matching and intersection parallel ratio matching,completing the target tracking of blocky debris;Finally,the double line counting algorithm is used to complete the quantity statistics of block shaped debris.(4)Design and testing of a block debris detection and detection system.Based on the above foundation,a detection scheme for block shaped debris detection system has been constructed,and a hardware system consisting of a visible light camera,industrial control computer,display screen,GPS/Beidou positioning module,communication module,etc.has been determined.The software system has achieved three functions:data acquisition,data analysis,and communication of detection results.The system was tested on a track inspection car and its feasibility was preliminarily verified.