| As a new intelligent inspection technology with high precision,high efficiency and a high degree of automation,robot vision positioning guidance is widely used in industry to meet the needs of automated industrial assembly.The combination of vision and industrial robots in automatic assembly systems has seen rapid development in recent years.As one of the most common types of electronic components on printed circuit boards,in-line electronic components,such as aluminium electrolytic capacitors,often have bent or offset pins and poor consistency,and often require manual assistance when performing parallel assembly of multi-axis holes,with large assembly errors and low success rates.This Master’s thesis takes a three-pin in-line capacitor and a corresponding PCB board as the experimental object for multi-axis hole assembly,and is oriented towards the key technologies for accurate gripping and precision assembly.The main tasks accomplished are:Firstly,a robot vision-based accurate gripping and precision assembly system was designed.A deep learning method for accurate grasping,multi-axis hole positioning and parallel assembly strategies are combined to design a precision assembly system for multi-axis hole direct insertion components.The system applies the target detection method to binocular vision to achieve fast and accurate grasping of capacitive components;the mechanical structure ensures that the shaft and hole cross-sections are parallel and the two monocular camera optical axes are perpendicular to the shaft and hole cross-sections,reducing the complex three-dimensional multi-axis hole assembly problem to two highly fixed two-dimensional plane assembly problems.The fast and precise visual positioning guidance algorithm replaces the complex and timeconsuming hole finding algorithm and contact force feedback control process,and improves the assembly success rate with the multi-axis hole parallel assembly strategy,finally realizing the non-contact multi-axis hole assembly with high accuracy and success rate.Secondly,a target detection and grasping task based on binocular vision and YOLOv7 network is implemented.A deep learning YOLOv7 network is used to identify capacitors in complex industrial environments,and after obtaining the coordinates of the centre of the capacitor anchor frame in the left and right cameras,its 3D world coordinates are calculated,and the robot is guided to grasp the workpiece and place it on the designated platform through a combination of demonstrative teaching and online recognition.The accuracy of the method is verified through experiments.Thirdly,a dual positioning algorithm and parallel assembly strategy for component centres and PCB assembly centres is proposed.A constrained RANSAC ellipse fitting algorithm is used to precisely locate the axis hole centre in combination with the CAD model;at the same time,an insertion guidance algorithm is developed that uses the multi-axis hole centre as the positioning anchor point and the average angular deviation of the multi-axis hole as the rotation basis,and guides the deployment by predicting the multi-axis hole assembly gap to improve the insertion success rate;finally,the assembly result is detected by an image difference algorithm to control the assembly quality.Fourthly,the system’s gripping success rate,assembly accuracy and assembly success rate were verified through experiments.The experimental data shows that the system has the advantages of high positioning assembly accuracy,high speed and high assembly success rate,achieving an overall inspection time of less than 100 ms for unit parts,a final assembly error of less than ±0.3 mm for a single axis and an assembly success rate of more than 98%. |
PDF Full Text Request