Research On Recognition Detection And Location Grab Method Of Butterfly Valve Body Based On Deep Learning

Butterfly valve is used to control the medium flow in low-pressure pipeline,which is widely used.At present,the butterfly valve assembly is completed by manual work,and the production efficiency is low.With the decrease of population dividend,the increase of workers' cost and the increase of manual operation,industrial robots are used instead of manual assembly.In order to improve the assembly efficiency and quality of butterfly valve,a method of valve body identification,detection and positioning grab based on deep learning is proposed.First of all,in the process of butterfly valve body identification,there are higher requirements for the accuracy and timeliness of butterfly valve body identification.A recognition and detection model based on improved YOLOv3 algorithm is proposed for the identification of butterfly valve body.The traditional recognition and detection algorithm is affected by the position and pose of the object,so the target detection algorithm based on deep learning has high generalization and stability.The butterfly valve body image information is collected by camera,and three deep learning algorithms,R-CNN,YOLOv3 and improved YOLOv3,are used to generate the butterfly valve body recognition and detection model for comparison.Experiments show that the recognition speed and accuracy of YOLOv3 algorithm is better than that of R-CNN algorithm,and the improved YOLOv3 algorithm improves the accuracy of the location of the recognition center.Secondly,in the process of positioning the grasping point of butterfly valve body,combined with the improved YOLOv3 recognition model,a method of positioning the grasping point of butterfly valve body is proposed,which can quickly determine the spatial position of the grasping point of butterfly valve body.The image information and depth information of butterfly valve body are collected by camera,and the distance between the valve body and the camera is obtained by the method of obtaining the distance between the valve body and the RGB-D camera;the conversion relationship between the pixel coordinate system and the world coordinate system is obtained by the coordinate conversion relationship between the valve body and the RGB-D camera;the conversion between the camera coordinate system and the robot base coordinate system is obtained by the hand eye calibration method of "eye in hand";The spatial position ofthe grasping point of the butterfly valve body in the industrial robot base coordinate system is obtained.Thirdly,in the process of grasping butterfly valve body,for the problem of low accuracy of grasping,a method of grasping butterfly valve body is proposed,which makes the robot grasping process more accurate.Combined with the method of body grasping point positioning and K-Means++ clustering algorithm,the posture of the valve body in the industrial robot base coordinate system is obtained,and then the posture of the gripper is obtained.Because the valve body of butterfly valve is an irregular rigid body,the claw can not be grasped at the grasping point of the valve body,so the grasping point of the valve body is offset to the valve seat direction for a certain distance as the actual grasping point of the valve body.This method optimizes the position and posture of the claw when grasping the valve body,improves the accuracy of grasping,and has good universality.Finally,the industrial robot is driven to grasp the butterfly valve body according to the position and actual grasping point coordinates of the valve body.The experiment results show that the method based on deep learning can complete the task of identification,detection and location grasping of butterfly valve body,and achieve the expected results.