Design Of Fastener Sorting System Based On Machine Vision

With the in-depth implementation of industry 4.0 and made in China 2025,robots play an increasingly important role in the development of industrial intelligent production.At present,there are many problems in the sorting work of large-scale mechanical parts production line,such as heavy tasks,low efficiency and high cost of manual sorting operations.Studying an intelligent sorting robot system based on machine vision technology to realize efficient and automatic sorting tasks is an important trend of industrial assembly line intelligence.The purpose of this paper is to develop a robot fastener sorting system based on machine vision,which can replace the mechanical sorting and manual sorting in traditional production line.The paper’s work can be separated into following five components:(1)On the basis of machine vision,this paper firstly completes the design of the overall scheme of robot sorting system and builds the hardware system.And then,the selection and installation methods of related hardware are determined and the communication methods between system components are discussed.On the VS software development platform,by calling the Halcon image processing library and the trained convolution neural network model interface,the development of the upper computer software system was completed.(2)This paper not only studies the forward and inverse kinematics of SCARA four-axis manipulator but also establishes the kinematics equations of the known joint angle and the robot’s end position.On this basis,Matlab software is used for kinematics simulation,SCARA robot is modeled and positive and negative solutions are verified,and then the kinematics analysis of SCARA robot is completed based on Adams software.In this paper,the camera internal parameters are calibrated by using the Halcon circular target punctuation method,and the calibration experiment of the hand eye system model is completed,and the pose conversion relationship between the camera and the manipulator is established.(3)The image preprocessing of the fastener and the target detection algorithm based on edge matching are studied,and the acquired images are processed by filtering,image enhancement and other preprocessing operations;on this basis,the edge extraction effect of Sobel,Prewitt and Canny on the image is compared;the dynamic workpiece is extracted from the image video sequence by the background difference method to complete the target detection,and then edge feature matching is used to complete the detection.Finally,Camshift algorithm is adopted to complete the dynamic tracking,which realizes the function of dynamic workpiece identification and tracking.(4)In order to overcome the disadvantages of the edge feature-based matching algorithm,such as poor anti-interference and low matching accuracy,this paper further studies the detection of the target by using deep learning method.An improved target detection algorithm based on Yolov3-tiny is proposed.Soft-NMS algorithm with Gauss weighted attenuation is used to replace NMS algorithm with hard decision to filter the redundant regression frame generated by prediction network.Not only 52 * 52 scale is added to the network structure to improve the detection accuracy of small target,but also but also mobilenetv2 basic structure block is used in the feature extraction network,which makes the network performance more excellent.(5)The experimental results show that the proposed fastener sorting system based on machine vision can achieve excellent performance in both target detection effect and sorting accuracy.The improved Yolov3-tiny target detection algorithm improves the detection ability of bolts,nuts,screws and washers,with the average accuracy increased from 0.813 to 0.839,nearly two percentage points higher;the recall rate increased from 0.804 to 0.821,increasing by nearly two percentage points;the average successful sorting rate of the system is maintained at about 90% by experiments,which makes the fastener sorting system based on machine vision have high accuracy and strong robustness.