基于机器视觉的电枢缺陷检测系统  

A System for Detecting Defects in Armature Based on Machine Vision

作　　者：王杨 王杰[1] 刘剑歌 方夏 李勇[1] 杨苗苗 WANG Yang;WANG Jie;LIU Jian-ge;FANG Xia;LI Yong;YANG Miao-miao(School of Mechanical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学机械工程学院,成都610065

出　　处：《组合机床与自动化加工技术》2021年第4期116-120,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：四川省重点研发项目(2019YFG0356,2021YFG0198,2021YFG0079);川大-泸州高校科研基金项目(2019CDLZ-24)。

摘　　要：为解决现阶段依靠肉眼进行电枢缺陷检测的效率低下和因为判断标准具有很大的主观性导致判别结果不稳定等问题,研究出了一种基于机器视觉的电枢缺陷检测系统。该系统可以采集到电枢不同角度的图像,并根据检测难度把每张电枢图像分为三个区域,分别由传统机器视觉算法、MobileNet-V3以及DenseNet神经网络模型对不同类型的缺陷同时进行检测。该系统所用到的数据集通过质检人员分类获得,可以保证标准的正确性。不同类型的缺陷分开检测可以提升系统的准确率,多区域同时检测可以提高系统的检测效率。实验表明,该系统在微型电枢的视觉缺陷检测上具有可靠的效果,能够应用于工业领域中并取代人工检测的方法。In order to solve the problems such as low efficiency of visual armature defect detection and unstable judgment result due to subjective judgment criteria,an armature defect detection system based on machine vision was developed.The system can collect images from different angles of the armature,and divide each armature image into three regions according to the detection difficulty,and simultaneously detect different types of defects by traditional machine vision algorithm,Mobilenet-V3 and DenseNet neural network model respectively.The data sets used in the system are classified by quality control personnel to ensure the correctness of the standards.Separate detection of different types of defects can improve the accuracy of the system,and simultaneous detection of multiple areas can improve the detection efficiency of the system.The experiment shows that the system has reliable effect on the visual defect detection of miniature armature,and can be applied to the industrial field and replace the manual detection method.

关 键 词：电枢 缺陷检测 传统机器视觉 协同神经网络 

分 类 号：TH165[机械工程—机械制造及自动化] TG65[金属学及工艺—金属切削加工及机床]