Research On Key Problems Of On-line Sorting System Based On Visual Servo

With the rapid development of 5G and other technologies,microchips have been widely used in various fields.In the process of packaging and manufacturing microchips,various defects will inevitably occur on the surface of the chip,which will directly affect the operation efficiency and life of the chip.The traditional manual visual detection method has been difficult to adapt to the high speed and high precision of microchip packaging and manufacturing.It has many advantages such as no damage,no contact and high stability.Although the chip defect detection technology based on machine vision has made good progress in chip printing characters,pin appearance size and position,the research on surface appearance defect detection and classification is still in the development stage.Therefore,this paper studies the key problems in the online surface defect sorting system of microchip based on visual servo,and proposes a complete theoretical system and system design scheme for online detection and sorting from image acquisition,image preprocessing,image segmentation to feature extraction and classification of surface defect.The main research contents of this paper are as follows:(1)Research on image preprocessing methods of surface defects of microchips.Firstly,aiming at the collected microchip images,a fast segmentation correction algorithm of chip defect images was proposed by combining two-dimensional wiener filter,Canny operator,Randon transform,bilinear interpolation and Hough operator respectively,so as to realize the rotation correction of chip images in the conveyor belt.Secondly,in order to improve the image resolution,highlight the defect features and protect the image edge information,an improved adaptive planar color interpolation algorithm is proposed.Experimental results show that the interpolation algorithm is superior to the comparison algorithm.(2)Research on multi-threshold segmentation method of surface defect image of microchip.In view of the characteristics of chip surface defect images,this paper firstly proposes a multi-threshold maximum entropy segmentation method based on the improved Elephant Herding algorithm(IEHO).This algorithm introduces the population competition and cooperation strategy and the improved acceptance solution strategy to accelerate the global exploration and local optimization of the image swarm.Then,in order to improve the standard puhuo algorithm in dealing with the image threshold segmentation more ability of global search ability and convergence,by introducing the improved adaptive balance search strategy and chaotic optimization strategy,put forward based on the improved algorithm of Moth Flame(IMFO)to multiple threshold maximum entropy image segmentation threshold image segmentation method.Experiments show that the IMFO algorithm is better than the comparison algorithm in population diversity,global search and local search,and convergence.(3)Study on feature extraction and classification and recognition of surface defects of microchips.Firstly,according to the actual appearance defect types of BGA chips and the actual classification requirements of chip images,four common defect types were selected in this paper for analysis and research.Secondly,25 kinds of geometric features,17 kinds of texture features and 3 kinds of gray features of all kinds of defective chips were extracted.Thirdly,in view of the miscellaneous information contained in the extracted 45 chip features,in order to eliminate the influence of irrelevant and redundant features as far as possible and reduce the dimension of the actual operation calculation,the principal component analysis(PCA)method was adopted to reduce the dimension of the 45 chip features.Fourthly,in view of the poor performance of SVM algorithm in solving multilevel classification,in order to improve the efficiency of chip classification,a support vector machine algorithm based on fireworks algorithm(FWA-SVM)was proposed to optimize the feature selection of defect feature samples based on the penalty factor and gaussian kernel parameter values of SVM,so as to generate the optimal feature subset.(4)Design,construction and implementation of visual servo system for surface defect detection of microchip.Firstly,the hardware function structure of the microchip detection system is designed,and the basic design functions of each hardware subsystem are described.Secondly,according to the requirements of the system for micro chip detection system software function structure design,put forward a kind of suitable for micro chip image online processing scheme,which can realize image acquisition from the chip,the chip image preprocessing,image threshold segmentation,feature extraction,dimension reduction,classification,judgment and the function of the mechanical arm sorting.Thirdly,the overall process of surface detection and defect classification and sorting in this system is described,including off-line training,online detection and classification and sorting.In the control of conveyor belt,programmable logic controller(PLC)is used to control the conveyor belt,and a matching visual servo microchip online sorting system monitoring software is developed to monitor it in real time.In terms of image acquisition and data transmission,the microscope camera is used to collect the images of the chip on the conveyor belt,and after image processing by industrial personal computer,the classification information is sent to STM32,and then transmitted to PAC controller,which controls the manipulator and pneumatic chuck to sort the target chip according to different classification paths.By selecting appropriate experimental methods and equipment,the actual servo system construction of this scheme was finally completed,and the experimental results achieved the expected requirements.(5)Online sorting experiment of surface defects of microchips.First of all,this paper introduces the microchip surface defect online index,the experimental environment of chip surface defect online detection and sorting.The microchip is then captured,sorted and monitored online.Finally,on the basis of realizing the complete process of microchip image rotation correction,interpolation amplification,threshold segmentation,defect area extraction,defect feature extraction and defect classification,the method of online defect detection and sorting proposed in this paper is experimentally verified and analyzed in terms of real time and practicability.The experiments show that it can satisfy the technical indexes of on-line detection and sorting of defective chips.