基于计算机视觉的轴承滚子表面缺陷在线检测系统

文生平, 周正军, 张啸言, 等

doi:10.12141/j.issn.1000-565X.190890

华南理工大学学报(自然科学版) >

2020 , Vol. 48 >Issue 10: 76 - 87

DOI: https://doi.org/10.12141/j.issn.1000-565X.190890

机械工程

基于计算机视觉的轴承滚子表面缺陷在线检测系统

展开

1．华南理工大学广东省高分子先进制造技术及装备重点实验室，广东广州 510640; 2．华南理工大学聚合物加工工程教育部重点实验室，广东广州 510640; 3．广州市井源机电设备有限公司，广东广州 511480

文生平(1966-)，男，博士，教授，主要从事智能控制与机器视觉等的研究。

收稿日期: 2019-12-09

修回日期: 2020-04-24

网络出版日期: 2020-09-14

基金资助

国家自然科学基金资助项目 (51973068); 国家重点研发计划项目 (2019YFC1908201)

收起

Online Detection System of Bearing Roller's Surface Defects Based on Computational Vision

Expand

1． Guangdong Advanced Polymer Manufacturing Technology and Equipment Key Laboratory，South China University of Technology，Guangzhou 510640，Guangdong，China; 2． Key Laboratory of Polymer Processing Engineering，the Ministry of Education，South China University of Technology，Guangzhou 510640，Guangdong，China; 3． Guangzhou Jingyuan Mechano-Electric Equipment Co．，Ltd．，Guangzhou 511480，Guangdong，China

文生平(1966-)，男，博士，教授，主要从事智能控制与机器视觉等的研究。

Received date: 2019-12-09

Revised date: 2020-04-24

Online published: 2020-09-14

Supported by

Supported by the National Natural Science Foundation of China (51973068) and the National Key R＆D Pro-gram of China (2019YFC1908201)

Fold

摘要

通过分析轴承滚子中最常见的几种表面缺陷类型，设计了针对性的缺陷检测算法，将传统计算机视觉方法与深度学习相结合，并采用改进的ＲetinaNet 模型，实现了轴承滚子的表面缺陷检测。实验结果表明: 文中方法的准确率达 95% 以上; 相较于传统的缺陷检测方法，文中方法在准确率、召回率与 F1-score 上均有一定提升。

关键词： 轴承; 表面缺陷; 在线检测; 深度学习; 卷积神经网络

本文引用格式

文生平, 周正军, 张啸言, 等 . 基于计算机视觉的轴承滚子表面缺陷在线检测系统[J]. 华南理工大学学报(自然科学版), 2020 , 48(10) : 76 -87 . DOI: 10.12141/j.issn.1000-565X.190890

Abstract

A reasonable detection algorithm for surface defects in bearing rollers was designed based on the analysis of some most common surface defect types． It combines the traditional computer vision method with deep learning and adopts the improved ＲetinaNet model to realize the surface defect detection of bearing rollers． The experimental results show that the accuracy of this method is more than 95%． As compared with the traditional defect detection method，the proposed detection algorithm can make improvement in accuracy，recall rate and F1-score．

Key words： bearing; surface defect; online detection; deep learning; convolution neural network

Options

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract