Research On Recognition And Grabbing Of Assemblies On Film Capacitor Production Line

Thin film capacitors have many outstanding advantages,such as high voltage and impact resistance,wide temperature range,good high frequency characteristics and stable frequency characteristics.They are widely used in the electronic industry.Especially with the rapid development of new energy industry in recent years,excellent characteristics of thin film capacitors can fully meet the requirements of new energy industry for capacitors.Therefore,thin film capacitors have an important society.It will have great significance and broad market application prospects.At present,the automation level of thin film capacitor production is not high.The main process of thin film capacitor production consists of winding,setting,wrapping and braiding(knitting plate),spraying gold,assembling quality inspection and so on.The object of the charging process of braiding and assembling is capacitor core and end cap respectively.This process often requires manpower to sort and feed,resulting in low production efficiency and quality not guaranteed.The tape and assembly manufacturers have an urgent need for automatic sorting and sorting feeding equipment,which forces film capacitor manufacturers and research units to accelerate the development of more intelligent and efficient equipment.With the rapid development of image processing technology,it has been widely used in automatic production line.Machine vision can guide industrial robots to complete the operation of each process instead of manual.In order to apply machine vision image processing technology to the identification of thin film capacitor assemblies,the robot is guided to sort and sort the thin film capacitor assemblies on the conveyor belt.In this paper,the recognition and positioning of target object and the calculation of deflection angle of multi-feature assembly are studied.The automatic recognition algorithm of multi-feature assembly parts under stacking condition is studied.A company parallel robot is selected to study the motion principle and control of the robot.The dynamic grasping experiment of film capacitor assembly parts on conveyor belt is designed,and the image is verified.The validity of physical algorithm and dynamic grabbing method is studied as follows:(1)Construction and introduction of hardware platform.According to the requirements of the test conditions,the combination of monocular vision and parallel robot is selected.Firstly,the camera imaging model and calibration principle are introduced.The conversion relationship between the pixels and the physical points in the world coordinate system is solved.The camera distortion parameters are obtained.The camera calibration experiment platform is built to complete the camera calibration.The relationship between user coordinate system and camera coordinate system is constructed,and the relationship between vision module and robot is constructed.At the same time,the communication between computer and robot is realized,so that the robot can obtain visual information.According to the experimental platform of machine vision,the guidance test of vision to robot is completed.(2)Target detection algorithm is studied.Aiming at the shape characteristics and placement status of capacitor core and end cap of two objects to be identified,the corresponding target detection algorithm is proposed.Firstly,image preprocessing methods such as OSTU threshold segmentation,morphological processing,filtering,image histogram equalization are used to study the Adaboost algorithm in machine learning and SSD target detection algorithm in depth learning.The principle and training process of the two algorithms are introduced.For capacitor cores with simple shape characteristics in non-stacking condition,Adaboost and SSD algorithms are directly used for recognition and detection,which can achieve high accuracy.Through the identification experiments of 60 randomly placed capacitor cores,it is found that both algorithms can achieve more than 99.5% accuracy.Aiming at the end cap with special shape feature in stacking condition,the two algorithms mentioned above can’t achieve the ideal detection accuracy directly,and there will be leakage detection for the covered object.In this paper,a detection algorithm for the workpiece with multiple features in stacking condition is proposed.This algorithm has high recognition accuracy for the end cap with stacking condition.Rate,the algorithm is used to identify 60 end cap images with stacked working conditions randomly,and the recognition accuracy can reach 99.8%.(3)The calculation method of target deflection angle is studied.In the process of tape and assembly quality inspection of thin film capacitors,it is required that the capacitor core and end cap should be placed in the correct posture.The deflection angle of the capacitor core and end cap relative to the correct position is required.In this paper,two angle calculation methods based on geometric feature points and template matching are proposed.Firstly,a maximum contour boundary object extraction method is introduced,which extracts the target area for deflection angle calculation based on the detection frame obtained by the target detection algorithm.The angle calculation method based on feature points is to calculate the deflection angle according to the position relationship between the feature points obtained on the capacitor core and the end cover.The principle of the algorithm based on template matching and the acceleration method used in the calculation process are analyzed.Two methods are used to calculate the angle of twelve randomly placed targets at the same time.Compared with the actual deflection angle,the average error of the angle calculation method based on feature points is 0.42 degrees,and the average error of the angle calculation method based on template matching is 0.15 degrees.(4)The proposed algorithm is verified by grabbing experiments.Target object is placed on the conveyor belt of uniform motion,and the target detection algorithm and angle calculation algorithm are dynamically grasped and verified.The actual coordinates of the target object are constantly changing.In order to realize the dynamic grasping of the robot,it is necessary to track the target.In this paper,displacement compensation is used to achieve the target tracking.The dynamic grasping process includes three stages: angle calculation of target recognition and coordinate conversion,information transmission and robot grasping.The displacement compensation of each stage is studied and calculated to achieve the initial target tracking.Setting the visual parameters of the robot,the speed value of the conveyor belt is the key.This paper uses image processing technology to complete the calculation of the speed of the conveyor belt.The grasping motion path of the robot is analyzed,and the recognition order of each target is rearranged by image processing method to ensure the orderliness of coordinate information sent to the robot.Finally,due to errors in the calculation of conveyor belt speed and robot grasping point,further adjustment of the final grasping point is needed.In order to achieve the goal of grasping,the end-effector should be redesigned.This paper uses sucker and stepping motor to achieve the goal of grasping and angle rotation.The successful rate of grabbing was 97.9% after 100 dynamic grabbing tests.