Research On Cleaning And Testing System Of Railway Freight Car Control Valve

With the continuous development of China’s rail transit field,it has become a beautiful scenery of China’s high-speed development.Automation and intelligent production has become a channel to improve industrial efficiency and quality.Air control valve in the field of rail transportation is the basis of the locomotive safety,need targeted on different types of valve body,valve hole and cavity cleaning,to meet the safety use standards of the air control valve,since most of the existing railway locomotive air control valve are cleaned and tested manually,Artifical visual fatigue can cause incomplete cleaning,inconsistent with detection standards,low efficiency and other.Under the intelligent cleaning and detection,similar problems can be effectively avoided.This article mainly to type 120 air control valve as an example for research,complete the robot automatic cleaning and inspection robot system design,and achieve type 120 air control valve main valve each valve hole,valve surface,cavity type identification research and its surrounding attached objects detection,and the simulation study on the path planning of the end-effector of the cleaning manipulator.The main research contents of this paper are as follows:1.Robot intelligent cleaning and detection system design.According to the railway freight car control valve system design,to ensure that the requirements of cleaning and detection can be realized,and taking type 120 air control valve as the research object,the corresponding cleaning process and cleaning time are specified,which lays a good foundation for the later identification and positioning.2.Research on Target Recognition and Positioning Method of Main Valve of Type 120 Air Control Valve.In this paper,the Yolov3 network structure layer is taken as the basis for improvement,and the valve body of type 120 air control valve is taken as an example to make a data set for experimental verification.The data set is divided into positive and negative samples for collection to prevent identification and positioning deviation.In the experiment,the weight information obtained after training was used to complete the identification experiment,and the position information,confidence and identification time of the corresponding target were outputted.The identification accuracy and identification time were verified by the verification set,and the position information of the identification box was provided for repositioning.3.Target recognition box repositioning and surface attachment detection.In order to improve the accuracy of the target position information,this paper relocates the target contour according to the information of the recognition box,preprocesses the image of the recognition box and detects the contour to get the edge information of the target.The draw-contours operator in Opencv is used for contour fitting to get the position of the fitting contour,and output the corresponding fitting size and position information.The number of the contour of the attachment is counted and the area is superimposed to complete the detection of the attachment and establish the detection and cleaning standard to ensure the smooth completion of the cleaning task.4.Research on path planning of end-effector of cleaning manipulator.In this paper,the heuristic function and the transfer direction of extension node are improved on the traditional D* algorithm,mix a variety of bounding box fusion in 3D simulation environment.The improved D* algorithm is used to carry out path planning for the end-effector of the manipulator arm,and the problems of too small corner and path redundancy in the 3D environment of the end-effector of the manipulator are fully solved,and verify the feasibility of path planning in actual production process.