Research On Surface Defect Detection Of Hollow Cup Motor Armature Based On Machine Vision

As the manufacturing model continues to develop toward intelligence,integration,and information,artificial intelligence technology and actual production continue to merge,and the market continues to put forward new requirements for product quality.In order to follow up market demand,product quality inspection technology based on artificial intelligence algorithms has become a hot spot for industry research and application.Among them,machine vision is the most in-depth research on product appearance inspection.Coreless motors are widely used in UAVs,intelligent robots and other industries due to their small moment of inertia and high energy conversion rate.Coreless armature is the core component of the motor,and its quality determines the service life of the motor.At present,the detection of hollow cup armature defects still relies on manual experience,which can no longer meet the development needs of manufacturing intelligence and integration.Aiming at the problems of the degree of automation and low efficiency of hollow cup armature appearance quality inspection,this paper carried out the application research of machine vision technology in hollow cup armature appearance defect inspection to improve the efficiency and accuracy of hollow cup armature appearance quality inspection.Push the intelligent construction of hollow cup armature manufacturing.The main research contents are as follows:(1)Build a hollow-cup armature appearance defect detection model based on the target detection algorithm.The Yolov4 network algorithm is used to achieve defect feature extraction and recognition classification.After comparison and verification with SSD,Inception V3 and other algorithms,the average recognition accuracy of this algorithm reaches 95.7%,and the detection time for a single image is 1.452 s,which can meet the appearance detection of hollow cup armature.Requirements for recognition accuracy and real-time performance.(2)Because of the many parameters of the Yolov4 model,it cannot be integrated on the low computing power equipment of the hollow cup armature production line.In this paper,model compression technology is used to study the weight of the Yolov4 algorithm.The Moblie Net V1 network is used to replace the backbone network of Yolov4 to ensure the accuracy and real-time performance of defect recognition,while achieving the lightweight of the Yolov4 model.(3)Based on the real-time inspection requirements of the hollow-cup armature intelligent manufacturing production line,a hollow-cup armature image acquisition platform was built,and a hollow-cup armature defect detection system was designed and developed,which can realize the function of real-time online detection of hollowcup armature defects.