Anti-loose Detection Of Bolt In Bullet Train Bogie Based On Deep Learning

In recent years,the mileage of China’s high-speed railway has been increasing year by year,and the technology related to the bullet train has been developing rapidly.The bogie is an indispensable part of the operating process of the bullet train.The bogie integrates bearing,guiding,vibration damping,traction and braking,which directly determines the operating stability and comfort of the train.However,the bogie can fall off and be damaged during operation.As a key component of the bullet train,the quality of the bogie is directly related to the normal operation of the bullet train and the safety of people’s lives and properties.Therefore,the bogie needs to be inspected in all aspects.At present,the bogie is mainly inspected manually,mainly for wheel pairs,bearings and other components.It is time-consuming,inefficient,and easy to be negligent,resulting in undetected defects.The number of bolt parts in the bogie is large,and the use of bolts to connect different structures can effectively enhance the structural stability of the bogie,which can ensure the smooth operation of the moving train and reduce the chance of safety accidents.In summary,this paper studies the automatic bogie detection technology and determines the detection scheme by combining the characteristics of the detection items,and the detection system is designed in order to achieve fast and accurate detection.The bolt loosening detection algorithm is studied and designed for the phenomenon of bolt loosening due to bolt loosening failure on the bogies of moving vehicles.1.The statistics of the inspection locations on the bogie were conducted,the inspection requirements and standards of the inspection items were summarized,and the 2D and 3D photo methods and inspection forms were determined.In order to improve the inspection efficiency and accuracy,a six-axis robot was used as the camera shooting carrier,and a bogie booster was designed to replace the manual pushing the bogie forward in order to realize automation.In the selection of the shooting light source,the projection type structure light source was chosen as the 2D photo light source;then the layout of the imaging method was designed,and the shooting device was designed as an adjustable device,which could be adjusted during the actual shooting.Finally,the layout of the entire detection system is installed.2.The situation of loose bolts and failure of anti-loosening measures is analyzed,and the method of locating the bolts and anti-loosening wires is studied,the detection data set of bolts and anti-loosening wires is established,and the deep learning target detection algorithm is used to locate and segment the bolts and anti-loosening wires.Based on the comparative analysis of existing target detection algorithms,the latest YOLOv5 s network has obvious superiority in the evaluation indexes such as detection speed,m AP value,accuracy rate and recall rate.It can achieve fast and accurate detection and localization.3.Detection of various bolt anti-loosening failures for the bolt and anti-loosening wire areas located by segmentation.(1)For the case of misalignment due to loosening of the white bolt locking marker wire,a positioning method based on color and shape position is proposed.The image to be detected is transferred into the HSV color space to exclude most of the interference information,and then the marker wire region is located by morphological processing and spatial location connectivity domain.After that,for the characteristics of the rectangular region,the detection of Hoff straight line is proposed,and the angular difference between two Hoff straight lines is calculated to determine whether the bolt is loose or not.In order to extract the Hough straight line accurately,two methods are proposed: Hough straight line detection based on binarized region and Hough straight line detection based on Canny edge detection;(2)For the 8-word tandem anti-loose wire orientation detection problem,due to the inconsistent shooting direction,it is proposed to square up the detection area with affine transformation,and then use deep learning target detection algorithm to determine the approximate positions of the bolt,anti-loose wire and bolt hole.(3)for the detection of the number of turns of anti-loosening wire with normal shooting angle,we propose to use YOLOv5 s network to segment the single turn,and then use the width of the whole section of anti-loosening wire to remove the width of the single turn to get the number of turns,and for the anti-loosening wire with variable shooting angle,use YOLOv5 s The number of turns is calculated by using the network to detect the inner connection rectangle marked inside the single turn.Finally,the bogie automatic detection software was developed and the bogie automatic detection experiments were conducted.The results showed that the system completed the detection in 27 minutes,the detection and positioning accuracy of bolts and locking wires reached 95%,and the false detection rate of bolt locking detection algorithm was less than 10%,which realized fast and accurate bogie bolt locking detection.