基于改进Faster R-CNN的绝缘轴承表面缺陷检测方法  

Detection Method for Surface Defects of Insulated Bearings Based on Improved Faster R-CNN

作　　者：高立明 贾书海[1] 张国龙 李勇 杨明奇[3] GAO Liming;JIA Shuhai;ZHANG Guolong;LI Yong;YANG Mingqi(School of Mechanical Engineering,Xi’an Jiaotong University,Xi’an 710049,China;Shanghai United Bearing Co.,Ltd.,Shanghai 200240,China;Luoyang Bearing Research Institute Co.,Ltd.,Luoyang 471039,China)

机构地区：[1]西安交通大学机械工程学院,西安710049 [2]上海联合滚动轴承有限公司,上海200240 [3]洛阳轴承研究所有限公司,河南洛阳471039

出　　处：《轴承》2023年第4期1-8,共8页Bearing

基　　金：国家重点研发计划资助项目(2020YFB2007900)。

摘　　要：喷涂陶瓷涂层的绝缘轴承外圈在经过磨削后,涂层表面会出现凹坑、夹杂和擦伤等缺陷,很可能成为涂层被轴电流击穿的安全隐患,目前此类缺陷大多采用人工目测,缺乏缺陷自动检测手段,现提出一种改进Faster R-CNN的绝缘轴承表面缺陷图像检测算法。首先,对绝缘轴承表面缺陷进行图像采集,并使用K-means++算法对缺陷数据集进行聚类,得到适用于此类缺陷的锚框;然后,用ROI Align替换ROI Pooling,避免ROI Pooling量化造成的定位误差;最后,在算法中加入在线难例挖掘策略以提高难检测样本的检测准确率。试验结果表明,改进Faster R-CNN网络对绝缘轴承表面缺陷的检测准确率达到91.2%,比Faster R-CNN网络提高了4.8%。After grinding of outer rings of insulated bearings sprayed with ceramic coating,the surface of coating will have defects such as pits,inclusions and scratches,which are likely to become a safety hazard of coating breakdown caused by shaft current.At present,most of these defects are visually detected and there is a lack of automatic defect detection methods.An image detection algorithm for surface defects of the bearings is proposed based on improved Faster R-CNN.Firstly,the images of surface defects of the bearings are acquired,and the data set of defects is clustered by using K-means++algorithm to obtain the anchor frame suitable for such defects.Then,the ROI Pooling is replaced by ROI Align to avoid the positioning errors caused by quantization of ROI Pooling.Finally,an online hard example mining is integrated into algorithm to improve the detection accuracy of difficult-to-detect samples.The test results show that the detection accuracy of surface defects of the bearings reaches 91.2%with improved Faster R-CNN network,which is 4.8%higher than that with Faster R-CNN network.

关 键 词：滚动轴承 绝缘轴承 陶瓷涂层 表面缺陷 自动检测装置 Faster R-CNN 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP216.1[自动化与计算机技术—检测技术与自动化装置]