Design And Research Of Cylindrical Lens Automatic Placing System Based On Machine Vision

At present,cylindrical lenses can be widely used in infrared lasers,optical instruments,holographic systems,and other fields.In the processing process of cylindrical lens,the placing process has a greater impact on the product yield.It is difficult to ensure that the production requirements can be met by the traditional manual placing workpiece clamping and placement,and the problem of long production time and low efficiency is one of the problems that needs to be solved in the cylindrical lens placing process.Therefore,a cylindrical lens automatic placing system based on machine vision was proposed in this paper,and a visual correction and positioning strategy based on U-NET network was constructed.Through structural design,software development and visual studies,the design and research of the placing system were carried out as follows:(1)The overall design of the cylindrical lens hardware platform was carried out.By analyzing the specific flow of the placing process,the overall scheme of the hardware platform was determined,the three-dimensional model of the cylindrical lens hardware platform was established by using Solidworks software,the parameter information of key components such as the Scara robot was studied,and the selection results of the vision hardware were calculated.(2)The operating software interface for cylindrical lenses was compiled and developed.Based on the hardware of the control system,the operation software and HMI interface were developed by the C# language,and the functional functions and visual interfaces suitable for operation were written according to the characteristics of the cylindrical lens placing process.The linear motion and acceleration control of the robot were studied,and the corresponding linear shape was adopted according to the actual movement situation,and the acceleration parameters were controlled by the "S" type speed curve,which can reduce the vibration and accuracy loss during operation.(3)A visual positioning correction strategy based on U-Net network was constructed.After calibrating the camera,the image information was enhanced with appropriate image pre-processing,and the coordinates of the suction and placement points of the workpiece under the robot coordinate system were obtained by using template matching,blob analysis and circle finding algorithm.Taking the original image of the workpiece stored in the pre-production process as the input data set,and the binarized picture generated by the corresponding high-precision workpiece segmentation data as the output data set,a U-Net image segmentation model was established to train and verify the cylindrical lens data.(4)The performance of many aspects of the placing system was compared and analyzed.Through the model training at different learning rates,it was found that when the learning rate was set to 0.0001,the loss value curve of the training set and the validation set decreased fastestly and converged to the minimum loss value with the fewest number of trainings.The visual positioning accuracy of the model was within one pixel relative to the high-precision template matching.Using the micrometer and stopper tools to measure at multiple points,the repeat positioning accuracy and surface type accuracy within 0.02 mm of the end of the robot were obtained.Finally,according to the data statistics,the eccentric pass rate of more than 35% and the product pass rate of more than 30% were relatively increased by the system placing,and the time consuming of a single disc operation was reduced to less than 1/5 compared with the manual placing method.