Research On Machine Vision Based Identification And Location Of Threaded Holes And Optimization Of Assembly Path

Screw locking is widely used in the assembly process of manufacturing industry as an assembly method,the traditional screw locking relies mainly on manual labor,which is inefficient and dependent on the experience of operators as well as easy to fatigue.The existing automatic screw locking device mainly used for teaching the positions of each screw hole and memorize the point position,while the operation is tedious,which is unable to identify and position the screw hole directly.Moreover,the assembly trajectory is not optimized and it has weak adaptation to different working conditions so it fails to achieve the real automation.In view of the existing problems in screw hole identification and positioning of screw locking equipment,based on machine vision and deep learning method,this thesis realizes the screw hole identification and positioning of the equipment and plans the path of the screw assembly.The main contents of this thesis are as follows:Firstly,a double camera vision detection system based on machine vision is designed according to the characteristics of the study object of this thesis,which uses two industrial cameras with different heights to operate separately.The general scheme of vision detection is designed,and the hardware and software of the system are introduced.Then,the thesis studies the circle detection algorithm,the Hough transform circle detection algorithm is selected as the method for the initial identification and positioning of the workpiece screw hole after comparison,the image pre-processing is implemented before the circle detection and the processing speed is improved by grayscale processing.Moreover,local histogram equalization is used to enhance the details of the image;finally,median filtering is used to remove image noise and improve the accuracy of detection.Thirdly,two industrial cameras were calibrated by the calibration method of Zhang Zhengyou camera in accordance with the actual need.When the coordinate transforms,the ant colony algorithm is used to optimize the image acquisition and assembly trajectory in the process of screw locking,and when the shortest path of image acquisition and screw assembly is found,optimize the path,then the industrial camera in the lower place collects the image according to the planned path,the collected image is used as the test sample image for deep learning of target detection.Finally,construct the Faster R-CNN target detection model to detect workpiece screw hole in combination with machine vision and deep learning algorithm and construct the data set of screw hole by myself.Using the migration learning to implement the training of the target detection model and detect the local images of screw hole collected by the camera in real time,and test the accuracy of the target detection model for the identification and positioning of the screw hole.