Research On Parts Grabbing System Based On Deep Convolution Neural Network

Industrial robots are the most typical mechatronics digital equipment and the supporting technology of intelligent manufacturing.They play an important role in industrial production and even social development.However,with the expansion of the application field of industrial robots,the requirement of intelligent industrial robots in manufacturing industry is getting higher and higher.The application of deep learning in the field of computer vision provides an important direction for intelligent robots.Therefore,it is a great significance for the research of robot parts grasping system based on deep learning.Firstly,this paper discusses the current popular deep learning algorithm.Referring to the application of deep learning in engineering at home and abroad,it is proposed that deep learning be applied to industrial robots to grasp common parts such as bolts,nuts,screws and gaskets.The main research focuses on the identification and location of four parts,the calibration algorithm of camera parameters,the estimation of the position and pose of parts grabbing,and the construction of the ultimate grabbing platform.The main contents of this paper are as follows:1)The target recognition algorithm VGG Net and the target detection algorithm Faster-RCNN are studied in depth.Then the two algorithms are optimized according to the requirements of the experiment in this paper,and the target recognition and target detection experiments are carried out using the self-built part data set.Four kinds of parts are identified and located.2)In order to find the mapping relationship between the two-dimensional position coordinates of the image and the three-dimensional coordinates of the robot,this paper deeply studies the calibration method of the camera's internal and external parameters,and calibrates the Kinect camera's internal and external parameters.Aiming at the feature that the parts picked up in this paper have the same depth in Z direction,this paper sets the Z-direction coordinates of the robot as a fixed value,which greatly simplifies the transformation of the two-dimensional coordinates of the image into three It is difficult to calculate.The experimental results show that this method can realize coordinate transformation within a certain tolerance range.3)In this paper,four kinds of parts are grasped by claw-hand and sucker.Aiming at the need of obtaining the sixth axis angle when the industrial robot claw-hand grasps the bolt correctly,the position and posture estimation of the bolt and bolt is proposed.Firstly,the data set of 10 angles of the parts is established,and then the recognition is carried out by using Google LeNet algorithm to obtain the closest angle within a certain tolerance range.The experimental results show that this method is feasible.In addition,for the grasp of gaskets,the sucker is used to exert offset distance in the Y direction of the center point of the gaskets to absorb,and for the grasp of nuts,the gripper is used to open the gripper at the center point to grasp.4)Aiming at the grasping of bolts,nuts,bolts and gaskets in this paper,a part grasping device for FANUC robot is designed,and a quick change joint is used to replace the clamping devices of different parts.Finally,this paper designs and builds an experimental platform of part grabbing system,and completes the overall experiment of part grabbing for industrial robots based on Faster-RCNN algorithm.The overall experimental results show that the part grabbing system can basically complete the experiment and meet the expected requirements.