作者：徐佳乐¹, 黄丹平¹, 廖世鹏², 甘芳吉³

作者单位：1. 四川轻化工大学机械工程学院，四川 宜宾 644000;
2. 中国科学院成都计算机应用研究所，四川 成都 610041;
3. 四川大学机械工程学院，四川 成都 610065

关键词：轴承表面缺陷;线阵相机扫描;图像重构;机器视觉;生成对抗网络

摘要：

针对目前轴承表面缺陷检测所面临问题，探求一种基于线扫描技术的轴承表面缺陷检测方法。由轴承表面特征，提出一种具有高信噪比的线扫描轴承表面信息采集系统。在此基础上，为解决浅凹坑、锈迹缺陷识别准确率低等问题，首先对所采集图像进行轴承区域提取，消除检测干扰，其次提出基于生成对抗的重构网络对采集到轴承信息进行图像重构，使其为无缺陷图像，最终通过残差方法提取和定位轴承表面缺陷。该算法重点对损失函数进行改进，采用L₁ loss与SSIM组合损失函数，提高图像的重构精度。实验表明：所提方法能有效识别各种缺陷，尤其是微小缺陷，综合识别率达到98.58%，满足实际工程精度要求。

Study on detection method of bearing surface defect based on line scanning technology
XU Jiale¹, HUANG Danping¹, LIAO Shipeng², GAN Fangji³
1. School of Mechanical Engineering, Sichuan University of Science & Engneering, Yibin 644000, China;
2. Chendu Institute of Computer Application, Chinese Academy of Sciences, Chengdu 610041, China;
3. School of Mechanical Engineering, Sichuan University, Chendu 610065, China
Abstract: Aiming at the problems of bearing surface defect detection, a method of bearing surface defect detection based on line scanning technology was proposed. Based on the characteristics of bearing surface, a bearing surface information acquisition system with high signal-to-noise ratio (SNR) was proposed. On this basis, in order to solve the recognition accuracy in shallow depressions and rust defect was low, firstly, bearing regions of collected images were extracted for eliminating interference of detection, secondly proposed reconstructing networks based on generative adversarial to be used for image reconstruction of the bearing information, make it to none-defect image, at the end, used residual method to extract and locate the bearing surface flaw. The algorithm focuses on the improvement of the loss function, used the combination of L₁ loss and SSIM loss function to improve the accuracy of image reconstruction. Experimental results show that the proposed method can effectively identify various defects, especially minor defect, and the comprehensive recognition ratio reaches 98.58%, which meets the requirement of actual engineering accuracy.
Keywords: bearing surface defect;line-array camera scanning;image reconstruction;machine vision;generative adversarial network
2022, 48(11):88-94  收稿日期: 2021-08-12;收到修改稿日期: 2021-10-27
基金项目: 四川省重点实验室项目（NJ2018-05）；自贡市科技局重点项目（2019YYJC12）；四川轻化工大学创新基金项目（y2020007）
作者简介: 徐佳乐（1996-），男，浙江余姚市人，硕士研究生，专业方向为智能检测、机器视觉、深度学习
参考文献
[1] 徐东, 徐永成. 单表面故障的滚动轴承系统非线性动力学研究[J]. 机械工程学报, 2010(21): 61-68
[2] 郝勇, 耿佩. 滚动轴承保持架缺陷的图像处理及模式识别方法研究[J]. 仪器仪表学报, 2019(9): 162-169
[3] 陈廉清, 袁红彬, 王龙山. SUSAN算子在微小轴承表面缺陷图像分割中的应用[J]. 光学技术, 2007(2): 305-307
[4] 张根源, 周泓. 存在点缺陷的深沟球轴承的动力学响应[J]. 浙江大学学报(工学版), 2009(8): 1497-1500
[5] 钱佳立, 陆惠宗, 袁巨龙, 等. 基于光学原理的轴承球体表面缺陷检测方法研究[J]. 表面技术, 2018(10): 309-314
[6] WANG H N, YANG J X, HU Z D. Current status and prospect of roller bearing surface defect detection[J]. Procedia Engineering, 2011(15): 4331-4336
[7] 董海江, 刘秀成, 何存富, 等 . 高频感应淬火45钢强度与硬化层深度的微磁无损检测[J]. 中国测试, 2021, 47(6): 6-12
[8] 项新建, 王乐乐, 曾航明. 基于机器视觉的推力轴承垫圈缺陷检测系统研究[J]. 中国测试, 2021, 47(2): 133-139
[9] 汪凤林, 周扬, 叶绿, 等. 基于机器视觉的飞轮齿圈缺陷和尺寸检测方法[J]. 中国测试, 2020, 46(5): 31-38
[10] 陈文达, 白瑞林. 基于机器视觉的轴承防尘盖表面缺陷检测[J]. 计算机工程与应用, 2014(6): 250-254
[11] 陈廉清, 崔治, 王龙山. 基于差影和模板匹配的微小轴承表面缺陷检测[J]. 中国机械工程, 2006(10): 1019-1022
[12] LIU B, YANG Y Q, WANG S S, et al. An automatic system for bearing surface tiny defect detection based on multi-angle illuminations[J]. Optik, 2020, 208: 164517
[13] 陈进. 基于机器视觉的轴承缺陷检测技术研究[D]. 杭州: 浙江大学, 2021.
[14] 徐同旭, 黄丹平. 新型光缆直径缺陷检测系统研究[J]. 现代制造工程, 2018(7): 116-123
[15] 安萌, 郑飂默. 一种改进Faster R-CNN的面料疵点检测方法[J]. 小型微型计算机系统, 2021(5): 1030-1033
[16] WEI X K, WEI D H,SUO D,et al. Multi-target defect identification for railway track line based on image processing and improved Yolov3 model[J]. IEEE ACCESS, 2020, 8: 61973-61988
[17] 张力, 黄丹平. 基于目标检测网络的轮对踏面缺陷检测方法[J]. 激光与光电子学进展, 2021, 58(4): 244-253