Automatic Cutting Study Of Air Conditioning Compressor Based On Vision Technologies

In recent years,the garbage recycling industry has developed in full swing,but the recycling and processing capacity of refrigeration appliances such as air conditioners is still insufficient.At present,as the core of air-conditioning and refrigeration equipment,the recovery and cutting method of compressor still rely on traditional manual cutting,which has low efficiency and is difficult to meet the needs of normal industrial production.Therefore,there is an urgent need in the industry for the complete,efficient and highly automated automatic cutting solution for air-conditioning compressors.Compressor edge key point detection and welding point detection are two key detection technologies in the cutting scheme.Firstly,thesis studies the key point detection technology of compressor edge.Thesis tries to use Hough line detection to obtain key points on the edge of the tank,but due to too much interference in the image,this method cannot achieve good detection results.Moreover,for the two kinds of tanks with high and short caps,thesis expects to obtain different detection results,but the Hough line detection can not meet the requirements.Therefore,thesis creatively designs a backtracking counting scanning algorithm to detect key points on the edge of the tank.This method directly utilizes the RGB features of the original image,introduces a counter and a backtracking mechanism,which can not only effectively suppress the interference in the image,but also effectively detect tanks with both high and short caps.Secondly,thesis studies the detection technology of compressor welding points.On the basis of YOLOv3,the YOLO-Dense Dark65 algorithm is designed to detect the welding points of the tank.The algorithm uses the Dense Convolutional Network to replace some of the convolutional layers in YOLOv3,and optimizes the original loss function of YOLOv3.By comparison,the recall rate,accuracy rate,m AP value and welding points detection success rate of YOLO-Dense Dark65 are significantly improved compared with YOLOv3、YOLOv5,which meets the actual needs of the industry.Thirdly,this paper designs a complete set of hand-eye cutting system that meets the needs of industrial production at the engineering site.The system is mainly composed of two parts: visual inspection system and assembly line cutting system.The computer vision algorithm is used to detect the position information of key points and welding points on the edge of the tank,and at the same time,a reasonable,scientific and efficient cutting path is planned,and then the cutting point is transmitted to the robotic arm to complete the cutting.Finally,four complete cutting schemes are designed and experimentally discussed in thesis: 1)four-piece separation method 2)true and false welding points classification method 3)full ring belt cutting method 4)door type compound cutting method.Generally speaking,these four schemes are a progressive relationship,but each has its own tradeoffs.All schemes must use the backtracking counting scanning algorithm for key points detection on the edge of the tank,and the last three schemes must use YOLODense Dark65 for welding points detection.The true and false welding points classification method detects the position of the welding points on the basis of the fourpiece separation method and classifies the true and false welding points of the tank,and adopts different cutting trajectories.The full ring belt cutting method uses YOLODense Dark65 to detect the connection hole and determine the lower end point,so as to solve the problem of the bottom of the tank touching the gun.Compared with other solutions,the door type compound cutting can divide the outer wall of the tank to the greatest extent and reduce the manual post-processing time.The cutting success rates of the four schemes are 62%,94.3%,97.3%,and 96.5%,respectively.The latter three schemes can meet the needs of industrial production.