Research On Intelligent Sorting Method Of Disordered Parts Based On Machine Learning

With the rapid development of the intelligent manufacturing industry,the use of robots instead of manual sorting on the assembly line has been widely used.There may be different types of parts placed disorderly on the assembly line.Currently,research on sorting disordered parts by industrial robots mostly focuses on traditional machine vision methods,often resulting in issues such as missed and false detection.In recent years,combining deep learning with robots has become a current research hotspot.Convolutional neural networks have good recognition performance and fast speed for different types of parts,but they are not precise enough for locating disordered parts.In this paper,a part detection method based on improved YOLOv4-tiny,combined with image processing technology,is proposed to realize the identification and precise positioning of disordered parts,and the conversion relationship between the pixel coordinate system and the base coordinate system of the robot is obtained by calibrating the internal and external parameters of the camera and the hand-eye calibration method,and the sorting task is completed by modeling the positive and inverse kinematics of the robot.The main work contents are as follows:(1)According to the needs of rapid identification of disordered parts,the advantages and disadvantages of the current mainstream detection algorithms are analyzed,and based on the YOLOv4-tiny part detection algorithm and improved,the CBAM attention mechanism is introduced into the enhanced feature extraction network part of YOLOv4-tiny,so that the feature extraction network pays attention to important feature areas.Aiming at the small amount of data of self-built part datasets,data augmentation is carried out through data augmentation,and K-means algorithm is used to cluster the data sets,and experiments show that the improved algorithm has good comprehensive detection ability.(2)The camera model is built and the calibration board is made,the checkerboard calibration method is used to solve the internal and external parameters and distortion coefficient of the camera model,the conversion relationship between the pixel coordinate system and the camera coordinate system is obtained,and the results are obtained through MATLAB software simulation,and the conversion relationship between the camera coordinate system and the robot base coordinate system is obtained by the method of eye in hand,and the pixel coordinates of the disordered parts are converted to the robot base coordinate system.(3)In view of the low positioning accuracy of disordered parts on the assembly line,further positioning is carried out on the basis of improving the predicted position of parts obtained by YOLOv4-tiny.By combining with Graham algorithm,the minimum circumscribed rectangle of the part contour is obtained after setting the optimal threshold area.The rotation angle of the rectangular frame and the center point of the long side are used as the grasping position of the execution end of the manipulator.(4)Combine the object detection module,vision module and robot grasping module to build an intelligent sorting platform.The forward and inverse kinematics model of the robot is established.When the robot grasps the position and pose information of the target part,the executive end of the manipulator can make selfadaptation adjustment to the grasping position and pose.On the basis of the above research,several sorting experiments have been carried out.The experimental results show that the intelligent sorting method proposed in this paper has high feasibility and improves the sorting ability of robots in complex scenes.