Research On Robotic Assembly System Based On Uncalibrated Visual Servoing

Due to the increase in high-rise buildings,the demand for elevators is growing.Many elevator companies began to implement large quantities of production lines.However,with the increase in labor costs year after year,enterprises are facing great pressure to survive.The introduction of automated equipment will change this situation.In order to meet the manufacturing requirements of the elevator workshop,reduce the labor cost and improve the speed and precision of the elevator door without riveting connection,a new national standard elevator door non-riveting connection system based on industrial robots was proposed and developed.The system has the advantages of high degree of automation,strong adaptability to different door panels,and high production efficiency.Robot teaching operations require accurate positioning of the target object.However,due to wear or equipment failure,the industrial site often makes the positioning inaccurate.In practice,the slight change of the camera position,focal length parameters or the camera lens distortion may result in inaccurate calibration model.This study uses the visual servo way to achieve the task of non-riveting connection,which exactly solves the problem mentioned above.With the industrial robot without calibration of the visual servo being as the object of study,the real-time vision controller and other issues of the non-calibration of the visual servo are discussed in this paper.The modified Jacobian matrix is fitted by improved genetic neural network.The main work is as follows:1.The new national standard elevator hall assembly process requirements is Introduced.The structural design of each station and the design of the robot vision servo system are described.Then the automatic control system of Omron CP1 H as the master PLC is constructed.A DeviceNet-based electrical communication system solution is established.2.Completed is the whole process of workpiece image processing.The different filtering methods and the edge detection operators are compared and analyzed.The feature extraction of the workpiece in complex background is realized by using the particle filter optimizing binarization image and Shi-Tomasi method.Then the input data of the subsequent neural network is obtained by image processing.3.The modeling and motion analysis of the non-riveting connecting robot are carried out to verify the multi-solution of the inverse kinematics of the robot.An improved GABP neural network is proposed,and an improved scheme of improved GABP neural network is proposed.4.After the comparison of the experiment results,it is obvious that the improved GABP has better network performance.When the improved visual controller is applied,the visual servo would receive a better performance as well,proving the reliability of the visual controller this work proposed.