Article(id=1244336749128823566, tenantId=1146029695717560320, journalId=1244311425741537314, issueId=1244336743298740932, articleNumber=null, orderNo=null, doi=10.16450/j.cnki.issn.1004-6801.2025.05.019, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1679241600000, receivedDateStr=2023-03-20, revisedDate=1692979200000, revisedDateStr=2023-08-26, acceptedDate=null, acceptedDateStr=null, onlineDate=1774602599650, onlineDateStr=2026-03-27, pubDate=1759248000000, pubDateStr=2025-10-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1774602599650, onlineIssueDateStr=2026-03-27, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1774602599650, creator=13701087609, updateTime=1774602599650, updator=13701087609, issue=Issue{id=1244336743298740932, tenantId=1146029695717560320, journalId=1244311425741537314, year='2025', volume='45', issue='5', pageStart='855', pageEnd='1056', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1774602598261, creator=13701087609, updateTime=1774603435030, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1244340253042000577, tenantId=1146029695717560320, journalId=1244311425741537314, issueId=1244336743298740932, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1244340253042000578, tenantId=1146029695717560320, journalId=1244311425741537314, issueId=1244336743298740932, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=1001, endPage=1007, ext={EN=ArticleExt(id=1244336749359510296, articleId=1244336749128823566, tenantId=1146029695717560320, journalId=1244311425741537314, language=EN, title=Wheelset Bearing Fault Detection Based on Multi‑resolution Siamese Network, columnId=1244336744728998604, journalTitle=Journal of Vibration,Measurement and Diagnosis, columnName=PAPER, runingTitle=null, highlight=null, articleAbstract=

In order to improve the fault detection performance of wheelset bearings under small sample image conditions,a machine vision inspection method based on a novel multi-resolution siamese neural network (MrSNN) is proposed for surface defect detection of wheelset bearings. First,the siamese neural network (SNN) is used as the basic model framework. A multi-resolution convolution fusion block (MrCFB) containing convolution kernels of different sizes and dilation factors is constructed to comprehensively extract the detailed features and contour features from images. Then,a dual attention mechanism combining channel and spatial information is adopted to recalibrate the multi-resolution feature weights,further enhancing the image feature extraction capability of the model. Finally,the algorithm is validated through the detection and analysis of four types of wheelset bearings images: normal,scratched,pitted and spalled. Experimental results show that the recognition rate for the three types of faulty images reaches 100%,the recognition rate for normal images is 95%,and the overall recognition accuracy is 98.75%. The recognition accuracy is superior to that of traditional SNN and YOLO-V5 models.

, correspAuthors=null, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=Feiyue DENG, Yan BI, Yongqiang LIU, Chunxue SONG, Rujiang HAO), CN=ArticleExt(id=1244336760453443730, articleId=1244336749128823566, tenantId=1146029695717560320, journalId=1244311425741537314, language=CN, title=基于多分辨率孪生网络的轮对轴承故障检测, columnId=1244336744909353678, journalTitle=振动、测试与诊断, columnName=论文, runingTitle=null, highlight=null, articleAbstract=

为了提高小样本图像条件下列车轮对轴承故障检测水平,提出了一种基于多分辨率孪生神经网络(multi⁃resolution siamese neural network,简称MrSNN)模型的列车轮对轴承表面缺陷机器视觉检测方法。首先,采用孪生神经网络(siamese neural network,简称SNN)为基础模型框架,构建了包含不同卷积核尺寸及不同膨胀因子大小的多分辨率卷积融合模块(multi‑resolution convolution fusion block,简称MrCFB)来综合提取图像中的细节特征与轮廓特征信息;其次,采用通道和空间的双重注意力机制重新标定多分辨率特征权重,进一步增强模型的图像特征提取能力;最后,通过对轮对轴承正常、划伤、凹痕及剥落4类图像进行检测分析,完成了算法验证。实验结果表明,轮对轴承3类故障图像识别率为100%,正常图像识别率为95%,总体识别准确率为98.75%,识别准确率优于传统SNN和YOLO‑V5等网络模型。

, correspAuthors=null, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=veosh2cCw2sUuaMiUgZB9w==, magXml=6PzxTV9VptwnaSd+PQPzlg==, pdfUrl=null, pdf=djNzL7k9A8qPlBg7mfMdKA==, pdfFileSize=3015384, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=t6vYQ+YOpmwkLhgZkl8KTw==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=dfKpfhfUj1m6zBPyjZQeZw==, mapNumber=null, authorCompany=null, fund=null, authors=

邓飞跃,男,1985年11月生,博士、副教授。主要研究方向为机械状态监测、信号处理、机器学习与故障诊断。曾发表《改进ILoG算子的故障检测方法》(《振动、测试与诊断》2020年第40卷第4期)等论文。 E-mail:

, authorsList=邓飞跃, 毕妍, 刘永强, 宋春雪, 郝如江)}, authors=[Author(id=1244351820206813233, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=dengfy@stdu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1244351820374585400, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820206813233, language=EN, stringName=Feiyue DENG, firstName=Feiyue, middleName=null, lastName=DENG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1244351820508803138, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820206813233, language=CN, stringName=邓飞跃, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1石家庄铁道大学机械工程学院 石家庄,050043, bio={"content":"

邓飞跃,男,1985年11月生,博士、副教授。主要研究方向为机械状态监测、信号处理、机器学习与故障诊断。曾发表《改进ILoG算子的故障检测方法》(《振动、测试与诊断》2020年第40卷第4期)等论文。 E-mail:

"}, bioImg=null, bioContent=

邓飞跃,男,1985年11月生,博士、副教授。主要研究方向为机械状态监测、信号处理、机器学习与故障诊断。曾发表《改进ILoG算子的故障检测方法》(《振动、测试与诊断》2020年第40卷第4期)等论文。 E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1244351818537480219, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=1., ext=[AuthorCompanyExt(id=1244351818545868828, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818554257437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1石家庄铁道大学机械工程学院 石家庄,050043)])]), Author(id=1244351820613660743, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1244351820722712654, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820613660743, language=EN, stringName=Yan BI, firstName=Yan, middleName=null, lastName=BI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1244351820814987351, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820613660743, language=CN, stringName=毕妍, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1石家庄铁道大学机械工程学院 石家庄,050043, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1244351818537480219, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=1., ext=[AuthorCompanyExt(id=1244351818545868828, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818554257437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1石家庄铁道大学机械工程学院 石家庄,050043)])]), Author(id=1244351820919844956, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1244351821070839909, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820919844956, language=EN, stringName=Yongqiang LIU, firstName=Yongqiang, middleName=null, lastName=LIU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1244351821230223470, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351820919844956, language=CN, stringName=刘永强, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1石家庄铁道大学机械工程学院 石家庄,050043, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1244351818537480219, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=1., ext=[AuthorCompanyExt(id=1244351818545868828, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818554257437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1石家庄铁道大学机械工程学院 石家庄,050043)])]), Author(id=1244351821330886773, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1244351821452521600, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351821330886773, language=EN, stringName=Chunxue SONG, firstName=Chunxue, middleName=null, lastName=SONG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2.CRRC Shijiazhuang Co.,Ltd. Shijiazhuang,050043,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1244351821607710856, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351821330886773, language=CN, stringName=宋春雪, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2中车石家庄车辆有限公司 石家庄,050043, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1244351818654920742, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=2., ext=[AuthorCompanyExt(id=1244351818663309351, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818654920742, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.CRRC Shijiazhuang Co.,Ltd. Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818667503655, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818654920742, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中车石家庄车辆有限公司 石家庄,050043)])]), Author(id=1244351821741928591, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1244351821821620373, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351821741928591, language=EN, stringName=Rujiang HAO, firstName=Rujiang, middleName=null, lastName=HAO, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1244351821960032415, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, authorId=1244351821741928591, language=CN, stringName=郝如江, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1石家庄铁道大学机械工程学院 石家庄,050043, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1244351818537480219, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=1., ext=[AuthorCompanyExt(id=1244351818545868828, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818554257437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1石家庄铁道大学机械工程学院 石家庄,050043)])])], keywords=[Keyword(id=1244351822169747627, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, orderNo=1, keyword=rolling bearing), Keyword(id=1244351822253633714, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, orderNo=2, keyword=fault diagnosis), Keyword(id=1244351822358491322, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, orderNo=3, keyword=neural network), Keyword(id=1244351822463348929, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, orderNo=4, keyword=image detection), Keyword(id=1244351822555623623, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, orderNo=5, keyword=multi-resolution feature), Keyword(id=1244351822647898322, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, orderNo=1, keyword=滚动轴承), Keyword(id=1244351822756950235, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, orderNo=2, keyword=故障诊断), Keyword(id=1244351822878585063, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, orderNo=3, keyword=神经网络), Keyword(id=1244351822983442672, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, orderNo=4, keyword=图像检测), Keyword(id=1244351823100883192, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, orderNo=5, keyword=多分辨率特征)], refs=[Reference(id=1244351829493002737, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2019, volume=126, issue=null, pageStart=137, pageEnd=160, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=HUANG W T, SUN H J, LUO J N, journalName=Mechanical Systems and Signal Processing, refType=null, unstructuredReference=HUANG W TSUN H JLUO J Net al. Periodic feature oriented adapted dictionary free OMP for rolling element bearing incipient fault diagnosis[J]. Mechanical Systems and Signal Processing2019126: 137-160., articleTitle=Periodic feature oriented adapted dictionary free OMP for rolling element bearing incipient fault diagnosis, refAbstract=null), Reference(id=1244351829623026173, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2023, volume=43, issue=2, pageStart=304, pageEnd=311, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=李永健, 宋浩, 李鹏, journalName=振动、测试与诊断, refType=null, unstructuredReference=李永健, 宋浩, 李鹏, . 改进离散熵在列车轴承损伤检测中的应用[J]. 振动、测试与诊断202343(2): 304-311., articleTitle=改进离散熵在列车轴承损伤检测中的应用, refAbstract=null), Reference(id=1244351829732078088, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2023, volume=43, issue=2, pageStart=304, pageEnd=311, url=null, language=null, rfNumber=2, rfOrder=2, authorNames=LI Yongjian, SONG Hao, LI Peng, journalName=Journal of Vibration, Measurement & Diagnosis, refType=null, unstructuredReference=LI YongjianSONG HaoLI Penget al. Application of improved dispersion entropy to fault detection of axle‑box bearing in train[J]. Journal of Vibration, Measurement & Diagnosis202343(2): 304-311.(in Chinese), articleTitle=Application of improved dispersion entropy to fault detection of axle‑box bearing in train, refAbstract=null), Reference(id=1244351829841129997, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2021, volume=9, issue=12, pageStart=312, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=YAN G, YU C Q, BAI Y, journalName=Machines, refType=null, unstructuredReference=YAN GYU C QBAI Y. A new hybrid ensemble deep learning model for train axle temperature short term forecasting[J]. Machines20219(12): 312., articleTitle=A new hybrid ensemble deep learning model for train axle temperature short term forecasting, refAbstract=null), Reference(id=1244351829929210386, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2017, volume=66, issue=4, pageStart=671, pageEnd=680, url=null, language=null, rfNumber=4, rfOrder=4, authorNames=ZHANG S B, HE Q B, ZHANG H B, journalName=IEEE Transactions on Instrumentation and Measurement, refType=null, unstructuredReference=ZHANG S BHE Q BZHANG H Bet al. Doppler correction using short-time MUSIC and angle interpolation resampling for wayside acoustic defective bearing diagnosis[J]. IEEE Transactions on Instrumentation and Measurement201766(4): 671-680., articleTitle=Doppler correction using short-time MUSIC and angle interpolation resampling for wayside acoustic defective bearing diagnosis, refAbstract=null), Reference(id=1244351830029873692, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=12, pageStart=169, pageEnd=172, url=null, language=null, rfNumber=5, rfOrder=5, authorNames=王恒迪, 李莎, 邓四二, journalName=机械设计与制造, refType=null, unstructuredReference=王恒迪, 李莎, 邓四二, . 轴承外圈侧面缺陷的视觉检测算法研究[J]. 机械设计与制造2017(12): 169-172., articleTitle=轴承外圈侧面缺陷的视觉检测算法研究, refAbstract=null), Reference(id=1244351830117954083, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=12, pageStart=169, pageEnd=172, url=null, language=null, rfNumber=5, rfOrder=6, authorNames=WANG Hengdi, LI Sha, DENG Sier, journalName=Machinery Design & Manufacture, refType=null, unstructuredReference=WANG HengdiLI ShaDENG Sieret al. Research on visual inspection algorithm of bearing outer ring side defects[J]. Machinery Design & Manufacture2017(12): 169-172.(in Chinese), articleTitle=Research on visual inspection algorithm of bearing outer ring side defects, refAbstract=null), Reference(id=1244351830222811687, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=2, pageStart=48, pageEnd=54, url=null, language=null, rfNumber=6, rfOrder=7, authorNames=陈硕, 林志敏, 吴岳彬, journalName=轴承, refType=null, unstructuredReference=陈硕, 林志敏, 吴岳彬, . 轴承套圈端面缺陷在线视觉检测的研究与实现[J]. 轴承2022(2): 48-54., articleTitle=轴承套圈端面缺陷在线视觉检测的研究与实现, refAbstract=null), Reference(id=1244351830319280687, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=2, pageStart=48, pageEnd=54, url=null, language=null, rfNumber=6, rfOrder=8, authorNames=CHEN Shuo, LIN Zhimin, WU Yuebin, journalName=Bearing, refType=null, unstructuredReference=CHEN ShuoLIN ZhiminWU Yuebinet al. Research and implementation of on-line visual inspection of defects on end face of bearing rings[J]. Bearing2022(2): 48-54.(in Chinese), articleTitle=Research and implementation of on-line visual inspection of defects on end face of bearing rings, refAbstract=null), Reference(id=1244351830440915510, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2022, volume=374, issue=4, pageStart=183, pageEnd=186, url=null, language=null, rfNumber=7, rfOrder=9, authorNames=石炜, 张袁祥, 李嘉楠, journalName=机械设计与制造, refType=null, unstructuredReference=石炜, 张袁祥, 李嘉楠. 列车滚子轴承表面缺陷机器视觉检测方法研究[J]. 机械设计与制造2022374(4): 183-186., articleTitle=列车滚子轴承表面缺陷机器视觉检测方法研究, refAbstract=null), Reference(id=1244351830533190203, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2022, volume=374, issue=4, pageStart=183, pageEnd=186, url=null, language=null, rfNumber=7, rfOrder=10, authorNames=SHI Wei, ZHANG Yuanxiang, LI Jianan, journalName=Machinery Design & Manufacture, refType=null, unstructuredReference=SHI WeiZHANG YuanxiangLI Jianan. Research on machine vision detection method for surface defects of train roller bearings[J]. Machinery Design & Manufacture2022374(4): 183-186.(in Chinese), articleTitle=Research on machine vision detection method for surface defects of train roller bearings, refAbstract=null), Reference(id=1244351830646436420, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2017, volume=45, issue=16, pageStart=160, pageEnd=164, url=null, language=null, rfNumber=8, rfOrder=11, authorNames=杨加东, 谢明, 王丽华, journalName=机床与液压, refType=null, unstructuredReference=杨加东, 谢明, 王丽华, . 基于BP神经网络的表面缺陷检测分类[J]. 机床与液压201745(16): 160-164., articleTitle=基于BP神经网络的表面缺陷检测分类, refAbstract=null), Reference(id=1244351830747099722, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2017, volume=45, issue=16, pageStart=160, pageEnd=164, url=null, language=null, rfNumber=8, rfOrder=12, authorNames=YANG Jiadong, XIE Ming, WANG Lihua, journalName=Machine Tool & Hydraulics, refType=null, unstructuredReference=YANG JiadongXIE MingWANG Lihuaet al. Surface defect detection and classification based on BP neural network[J]. Machine Tool & Hydraulics201745(16): 160-164.(in Chinese), articleTitle=Surface defect detection and classification based on BP neural network, refAbstract=null), Reference(id=1244351830847763026, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2019, volume=16, issue=4, pageStart=1836, pageEnd=1847, url=null, language=null, rfNumber=9, rfOrder=13, authorNames=XIE Q, LI D W, XU J X, journalName=IEEE Transactions on Automation Science and Engineering, refType=null, unstructuredReference=XIE QLI D WXU J Xet al. Automatic detection and classification of sewer defects via hierarchical deep learning[J]. IEEE Transactions on Automation Science and Engineering201916(4): 1836-1847., articleTitle=Automatic detection and classification of sewer defects via hierarchical deep learning, refAbstract=null), Reference(id=1244351830948426329, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2018, volume=94, issue=9, pageStart=3465, pageEnd=3471, url=null, language=null, rfNumber=10, rfOrder=14, authorNames=WANG T, CHEN Y, QIAO M N, journalName=The International Journal of Advanced Manufacturing Technology, refType=null, unstructuredReference=WANG TCHEN YQIAO M Net al. A fast and robust convolutional neural network-based defect detection model in product quality control[J]. The International Journal of Advanced Manufacturing Technology201894(9): 3465-3471., articleTitle=A fast and robust convolutional neural network-based defect detection model in product quality control, refAbstract=null), Reference(id=1244351831032312413, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2019, volume=157, issue=null, pageStart=417, pageEnd=426, url=null, language=null, rfNumber=11, rfOrder=15, authorNames=TIAN Y N, YANG G D, WANG Z, journalName=Computers and Electronics in Agriculture, refType=null, unstructuredReference=TIAN Y NYANG G DWANG Zet al. Apple detection during different growth stages in orchards using the improved YOLO-V3 model[J]. Computers and Electronics in Agriculture2019157: 417-426., articleTitle=Apple detection during different growth stages in orchards using the improved YOLO-V3 model, refAbstract=null), Reference(id=1244351831124587107, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=16249, pageEnd=16259, url=null, language=null, rfNumber=12, rfOrder=16, authorNames=YANG G L, TANG H, DING M L, journalName=null, refType=null, unstructuredReference=YANG G LTANG HDING M Let al. Transformer-based attention networks for continuous pixel-wise prediction[C]∥2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada: IEEE, 2021: 16249-16259., articleTitle=Transformer-based attention networks for continuous pixel-wise prediction, refAbstract=null), Reference(id=1244351831204278885, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2018, volume=110, issue=null, pageStart=349, pageEnd=367, url=null, language=null, rfNumber=13, rfOrder=17, authorNames=JIA F, LEI Y G, LU N, journalName=Mechanical Systems and Signal Processing, refType=null, unstructuredReference=JIA FLEI Y GLU Net al. Deep normalized convolutional neural network for imbalanced fault classification of machinery and its understanding via visualization[J]. Mechanical Systems and Signal Processing2018110: 349-367., articleTitle=Deep normalized convolutional neural network for imbalanced fault classification of machinery and its understanding via visualization, refAbstract=null), Reference(id=1244351831317525100, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=2832, pageEnd=2836, url=null, language=null, rfNumber=14, rfOrder=18, authorNames=ZHANG C, LIU W, MA H D, journalName=null, refType=null, unstructuredReference=ZHANG CLIU WMA H Det al. Siamese neural network based gait recognition for human identification[C]∥2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Shanghai, China: IEEE, 2016: 2832-2836., articleTitle=Siamese neural network based gait recognition for human identification, refAbstract=null), Reference(id=1244351831430771314, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2020, volume=3, issue=null, pageStart=48, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=19, authorNames=LI M D, CHANG K, BEARCE B, journalName=NPJ Digital Medicine, refType=null, unstructuredReference=LI M DCHANG KBEARCE Bet al. Siamese neural networks for continuous disease severity evaluation and change detection in medical imaging[J]. NPJ Digital Medicine20203: 48., articleTitle=Siamese neural networks for continuous disease severity evaluation and change detection in medical imaging, refAbstract=null), Reference(id=1244351831539823224, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2022, volume=75, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=20, authorNames=FU R R, WANG Y D, JIA C C, journalName=Biomedical Signal Processing and Control, refType=null, unstructuredReference=FU R RWANG Y DJIA C C. Data augmentation for cross-subject EEG features using Siamese neural network[J]. Biomedical Signal Processing and Control202275: 103614., articleTitle=Data augmentation for cross-subject EEG features using Siamese neural network, refAbstract=null), Reference(id=1244351831632097918, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2021, volume=14, issue=6, pageStart=3984, pageEnd=3997, url=null, language=null, rfNumber=17, rfOrder=21, authorNames=LI, Z, CHEN, H, LIN, K, journalName=Peer-to-Peer Networking and Applications, refType=null, unstructuredReference=LI, ZCHEN, HLIN, Ket al. From edge data to recommendation: A double attention-based deformable convolutional network[J]. Peer-to-Peer Networking and Applications202114(6): 3984-3997., articleTitle=From edge data to recommendation: A double attention-based deformable convolutional network, refAbstract=null), Reference(id=1244351831749538437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2023, volume=275, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=22, authorNames=CHEN L J, YAO H D, FU J Y, journalName=Engineering Structures, refType=null, unstructuredReference=CHEN L JYAO H DFU J Yet al. The classification and localization of crack using lightweight convolutional neural network with CBAM[J]. Engineering Structures2023275: 115291., articleTitle=The classification and localization of crack using lightweight convolutional neural network with CBAM, refAbstract=null), Reference(id=1244351831837618827, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2019, volume=30, issue=null, pageStart=41, pageEnd=47, url=null, language=null, rfNumber=19, rfOrder=23, authorNames=LU S Y, LU Z H, ZHANG Y D, journalName=Journal of Computational Science, refType=null, unstructuredReference=LU S YLU Z HZHANG Y D. Pathological brain detection based on AlexNet and transfer learning[J]. Journal of Computational Science201930: 41-47., articleTitle=Pathological brain detection based on AlexNet and transfer learning, refAbstract=null), Reference(id=1244351831925699217, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=770, pageEnd=778, url=null, language=null, rfNumber=20, rfOrder=24, authorNames=HE K M, ZHANG X Y, REN S Q, journalName=null, refType=null, unstructuredReference=HE K MZHANG X YREN S Qet al. Deep residual learning for image recognition[C]∥2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 770-778., articleTitle=Deep residual learning for image recognition, refAbstract=null), Reference(id=1244351832017973913, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, doi=null, pmid=null, pmcid=null, year=2021, volume=16, issue=3, pageStart=5390, pageEnd=5406, url=null, language=null, rfNumber=21, rfOrder=25, authorNames=FANG Y M, GUO X X, CHEN K, journalName=BioResources, refType=null, unstructuredReference=FANG Y MGUO X XCHEN Ket al. Accurate and automated detection of surface knots on sawn timbers using YOLO-V5 model[J]. BioResources202116(3): 5390-5406., articleTitle=Accurate and automated detection of surface knots on sawn timbers using YOLO-V5 model, refAbstract=null)], funds=[Fund(id=1244351829245538783, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, awardId=12272243, language=CN, fundingSource=国家自然科学基金资助项目(12272243), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1244351818537480219, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=1., ext=[AuthorCompanyExt(id=1244351818545868828, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818554257437, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818537480219, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1石家庄铁道大学机械工程学院 石家庄,050043)]), AuthorCompany(id=1244351818654920742, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, xref=2., ext=[AuthorCompanyExt(id=1244351818663309351, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818654920742, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2.CRRC Shijiazhuang Co.,Ltd. Shijiazhuang,050043,China), AuthorCompanyExt(id=1244351818667503655, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, companyId=1244351818654920742, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中车石家庄车辆有限公司 石家庄,050043)])], figs=[ArticleFig(id=1244351824883462408, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.1, caption=Framework of the SNN, figureFileSmall=K0yWqMGkLKGvVJhbrUQkIw==, figureFileBig=lULoe/okEI3nYFxh4wwuwA==, tableContent=null), ArticleFig(id=1244351825017680147, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图1, caption=孪生神经网络框架, figureFileSmall=K0yWqMGkLKGvVJhbrUQkIw==, figureFileBig=lULoe/okEI3nYFxh4wwuwA==, tableContent=null), ArticleFig(id=1244351825135120667, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.2, caption=Convolution operations with different convolution kernel sizes, figureFileSmall=DaWYyCo29Vzis4ntivrbNg==, figureFileBig=LuWNmY7v+4mko9TVzs0OEQ==, tableContent=null), ArticleFig(id=1244351825265144103, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图2, caption=不同卷积核尺寸的卷积操作, figureFileSmall=DaWYyCo29Vzis4ntivrbNg==, figureFileBig=LuWNmY7v+4mko9TVzs0OEQ==, tableContent=null), ArticleFig(id=1244351825340641580, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.3, caption=Convolution operation using different dilation factors, figureFileSmall=Fk1TO/JJd0BWn4CPojOKlg==, figureFileBig=jjjWBYGFbefuWQOMl/RfFA==, tableContent=null), ArticleFig(id=1244351825462276402, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图3, caption=不同膨胀因子的卷积操作, figureFileSmall=Fk1TO/JJd0BWn4CPojOKlg==, figureFileBig=jjjWBYGFbefuWQOMl/RfFA==, tableContent=null), ArticleFig(id=1244351825520996666, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.4, caption=Multi-resolution convolution fusion block, figureFileSmall=MYMfKULXdIzWGCsBuY0UIw==, figureFileBig=ezjRFt4WCqX94WR3LI8MTA==, tableContent=null), ArticleFig(id=1244351825596494144, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图4, caption=多分辨率卷积融合模块, figureFileSmall=MYMfKULXdIzWGCsBuY0UIw==, figureFileBig=ezjRFt4WCqX94WR3LI8MTA==, tableContent=null), ArticleFig(id=1244351825697157445, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.5, caption=Double attention mechanism module, figureFileSmall=XZ2N0BJ2BRWcXjqyd33/eQ==, figureFileBig=5Npc5mBNaA69J2ucOGYVVQ==, tableContent=null), ArticleFig(id=1244351825785237838, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图5, caption=双重注意力机制模块, figureFileSmall=XZ2N0BJ2BRWcXjqyd33/eQ==, figureFileBig=5Npc5mBNaA69J2ucOGYVVQ==, tableContent=null), ArticleFig(id=1244351825894289749, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.6, caption=Images of some wheelset bearings, figureFileSmall=AB0PWKhWO8v0Sgr4ZQ1MJA==, figureFileBig=1TjTbEo1WzXQ3Z2UJfjhKQ==, tableContent=null), ArticleFig(id=1244351826032701791, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图6, caption=部分轮对轴承图像, figureFileSmall=AB0PWKhWO8v0Sgr4ZQ1MJA==, figureFileBig=1TjTbEo1WzXQ3Z2UJfjhKQ==, tableContent=null), ArticleFig(id=1244351826104004962, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.7, caption=Feature extraction network based on feature enhancement, figureFileSmall=wpO/io76T5fKPTgxljjMNg==, figureFileBig=B0OeDnOP37Qf9xtVC3FUqA==, tableContent=null), ArticleFig(id=1244351826204668268, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图7, caption=基于特征增强的特征提取网络, figureFileSmall=wpO/io76T5fKPTgxljjMNg==, figureFileBig=B0OeDnOP37Qf9xtVC3FUqA==, tableContent=null), ArticleFig(id=1244351826326303091, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.8, caption=Iterative process of loss function, figureFileSmall=O4i9qO5/2YDMSrkV/4nCOQ==, figureFileBig=Yfj/uLKZ+kjUW1tUByuvOA==, tableContent=null), ArticleFig(id=1244351826397606264, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图8, caption=所提网络的损失值迭代过程, figureFileSmall=O4i9qO5/2YDMSrkV/4nCOQ==, figureFileBig=Yfj/uLKZ+kjUW1tUByuvOA==, tableContent=null), ArticleFig(id=1244351826510852480, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.9, caption=The confusion matrix of test result, figureFileSmall=PI8qQQIutWIFsddbHsLPrA==, figureFileBig=Bl+dOqBdP0h9THp4DWOoug==, tableContent=null), ArticleFig(id=1244351826615710085, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图9, caption=测试结果混淆矩阵, figureFileSmall=PI8qQQIutWIFsddbHsLPrA==, figureFileBig=Bl+dOqBdP0h9THp4DWOoug==, tableContent=null), ArticleFig(id=1244351826716373389, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.10, caption=Influence result of different batch sizes, figureFileSmall=hoa+qZqOFHXdlWlEm4eWUg==, figureFileBig=6tLuNSbcmprF+Vzpgpf5cw==, tableContent=null), ArticleFig(id=1244351826791870869, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图10, caption=不同批量大小的影响结果, figureFileSmall=hoa+qZqOFHXdlWlEm4eWUg==, figureFileBig=6tLuNSbcmprF+Vzpgpf5cw==, tableContent=null), ArticleFig(id=1244351826879951261, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.11, caption=Influence result of different block numbers, figureFileSmall=1l1pSR2HfFzz7GB1TxpXwg==, figureFileBig=jsmJmIwlYddVTJPCvg1kRw==, tableContent=null), ArticleFig(id=1244351827005780389, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图11, caption=不同模块数量的影响结果, figureFileSmall=1l1pSR2HfFzz7GB1TxpXwg==, figureFileBig=jsmJmIwlYddVTJPCvg1kRw==, tableContent=null), ArticleFig(id=1244351827093860782, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.12, caption=Influence result of different thresholds, figureFileSmall=vJ5CpQvxRGB3fYOz/4MH3A==, figureFileBig=Utb2zxPpm6cxum8LdtGgYg==, tableContent=null), ArticleFig(id=1244351827177746866, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图12, caption=不同阈值的影响结果, figureFileSmall=vJ5CpQvxRGB3fYOz/4MH3A==, figureFileBig=Utb2zxPpm6cxum8LdtGgYg==, tableContent=null), ArticleFig(id=1244351827324547515, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Fig.13, caption=Comparison results of recognition accuracy, figureFileSmall=SX18f4cnxOwRXPCO1HsJ5w==, figureFileBig=nuPGB4RXFy3XNUx6WT/lgg==, tableContent=null), ArticleFig(id=1244351827441988034, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=图13, caption=识别准确率结果对比, figureFileSmall=SX18f4cnxOwRXPCO1HsJ5w==, figureFileBig=nuPGB4RXFy3XNUx6WT/lgg==, tableContent=null), ArticleFig(id=1244351827525874119, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=EN, label=Tab.1, caption=

Defect recognition accuracy of the proposed network model

, figureFileSmall=null, figureFileBig=null, tableContent=
缺陷类型正常划伤凹痕剥落
正常95500
划伤010000
凹痕001000
剥落000100
), ArticleFig(id=1244351827634926033, tenantId=1146029695717560320, journalId=1244311425741537314, articleId=1244336749128823566, language=CN, label=表1, caption=

所提网络模型缺陷识别准确率

, figureFileSmall=null, figureFileBig=null, tableContent=
缺陷类型正常划伤凹痕剥落
正常95500
划伤010000
凹痕001000
剥落000100
)], attaches=null, journal=Journal(id=1244311337640177711, delFlag=0, nameCn=振动、测试与诊断, nameEn=Journal of Vibration,Measurement and Diagnosis, nameHistory1=null, nameHistory2=null, issn=1004-6801, eissn=null, cn=32-1361/V, coden=null, periodic=1, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=95p/itvXy4EP7rhNXfiFtg==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Vibration,Measurement and Diagnosis, journalRemark=null, publicationField=null, createdTime=1774596541081, updatedTime=1774596898872, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Engineering, subjectName=null, subjectCodeEn=Engineering, subjectNameEn=null, picCn=95p/itvXy4EP7rhNXfiFtg==, picEn=xUyZxfSFIgBcHbcc5pTNzg==, jcr=null, cjcr=null, exts=[JournalExt(id=1244312838404157841, language=CN, name=振动、测试与诊断, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1774596898890, updatedTime=1774596898890, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/author/login, submissionEditorUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/editor/login, submissionReviewUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1244312838450295186, language=EN, name=Journal of Vibration,Measurement and Diagnosis, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1774596898901, updatedTime=1774596898901, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/author/login, submissionEditorUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/editor/login, submissionReviewUrl=https://zdcs.nuaa.edu.cn/zdcsyzd/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1244311425741537314, websiteList=[Website(id=1244317743663661658, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1244311425741537314, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/zdcsyzd/CN, language=CN, createTime=1774598068395, createBy=18614031015, updateTime=1774598091436, updateBy=18614031015, name=振动、测试与诊断-中文, tplId=1146099689490845704, title=振动、测试与诊断, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1244318053069074825, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=articleTextType, value=kx, createTime=1774598142163, updateTime=1774598142163, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053052297606, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=banner, value=null, createTime=1774598142159, updateTime=1774598142159, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053090046348, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=grayFlag, value=0, createTime=1774598142168, updateTime=1774598142168, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053048103301, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=logo, value=https://castjournals.cast.org.cn/joweb/zdcsyzd/CN/file/pic?fileId=puE8pfA1j6xJq1REk82Yig==, createTime=1774598142158, updateTime=1774598142158, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053102629262, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=minRunFlag, value=0, createTime=1774598142171, updateTime=1774598142171, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053064880520, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zdcsyzd/CN/file/pic, createTime=1774598142162, updateTime=1774598142162, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053094240653, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=silenceFlag, value=0, createTime=1774598142169, updateTime=1774598142169, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053056491911, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1774598142160, updateTime=1774598142160, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053077463434, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=themeColor, value=null, createTime=1774598142165, updateTime=1774598142165, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318053081657739, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743663661658, code=themeStyle, value=null, createTime=1774598142166, updateTime=1774598142166, creator=18614031015, updator=18614031015)]), Website(id=1244317743730770523, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1244311425741537314, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/zdcsyzd/EN, language=EN, createTime=1774598068411, createBy=18614031015, updateTime=1774598107332, updateBy=18614031015, name=振动、测试与诊断-英文, tplId=1146101810881728533, title=Journal of Vibration,Measurement and Diagnosis, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1244318079505777348, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=articleTextType, value=kx, createTime=1774598148466, updateTime=1774598148466, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079484805825, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=banner, value=null, createTime=1774598148461, updateTime=1774598148461, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079526748871, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=grayFlag, value=0, createTime=1774598148471, updateTime=1774598148471, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079476417216, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=logo, value=https://castjournals.cast.org.cn/joweb/zdcsyzd/EN/file/pic?fileId=puE8pfA1j6xJq1REk82Yig==, createTime=1774598148459, updateTime=1774598148459, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079539331785, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=minRunFlag, value=0, createTime=1774598148474, updateTime=1774598148474, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079497388739, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zdcsyzd/EN/file/pic, createTime=1774598148464, updateTime=1774598148464, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079530943176, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=silenceFlag, value=0, createTime=1774598148472, updateTime=1774598148472, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079493194434, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1774598148463, updateTime=1774598148463, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079514165957, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=themeColor, value=null, createTime=1774598148468, updateTime=1774598148468, creator=18614031015, updator=18614031015), WebsiteProps(id=1244318079518360262, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1244317743730770523, code=themeStyle, value=null, createTime=1774598148469, updateTime=1774598148469, creator=18614031015, updator=18614031015)])], journalTitle=振动、测试与诊断, weixinUrl=null, journalUrl=https://zdcs.nuaa.edu.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Vibration,Measurement and Diagnosis, journalPhotoCn=95p/itvXy4EP7rhNXfiFtg==, journalPhotoEn=xUyZxfSFIgBcHbcc5pTNzg==, journalFirstLetter=J, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/zdcsyzd/CN/10.16450/j.cnki.issn.1004-6801.2025.05.019, detailUrlEn=https://castjournals.cast.org.cn/joweb/zdcsyzd/EN/10.16450/j.cnki.issn.1004-6801.2025.05.019, pdfUrlCn=https://castjournals.cast.org.cn/joweb/zdcsyzd/CN/PDF/10.16450/j.cnki.issn.1004-6801.2025.05.019, pdfUrlEn=https://castjournals.cast.org.cn/joweb/zdcsyzd/EN/PDF/10.16450/j.cnki.issn.1004-6801.2025.05.019, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于多分辨率孪生网络的轮对轴承故障检测
收藏切换
PDF下载
邓飞跃 1 , 毕妍 1 , 刘永强 1 , 宋春雪 2 , 郝如江 1
振动、测试与诊断 | 论文 2025,45(5): 1001-1007
收起
收藏切换
振动、测试与诊断 | 论文 2025, 45(5): 1001-1007
基于多分辨率孪生网络的轮对轴承故障检测
全屏
邓飞跃1 , 毕妍1, 刘永强1, 宋春雪2, 郝如江1
作者信息
  • 1石家庄铁道大学机械工程学院 石家庄,050043
  • 2中车石家庄车辆有限公司 石家庄,050043
  • 邓飞跃,男,1985年11月生,博士、副教授。主要研究方向为机械状态监测、信号处理、机器学习与故障诊断。曾发表《改进ILoG算子的故障检测方法》(《振动、测试与诊断》2020年第40卷第4期)等论文。 E-mail:

Wheelset Bearing Fault Detection Based on Multi‑resolution Siamese Network
Feiyue DENG1 , Yan BI1, Yongqiang LIU1, Chunxue SONG2, Rujiang HAO1
Affiliations
  • 1.School of Mechanical Engineering,Shijiazhuang Tiedao University Shijiazhuang,050043,China
  • 2.CRRC Shijiazhuang Co.,Ltd. Shijiazhuang,050043,China
出版时间: 2025-10-01 doi: 10.16450/j.cnki.issn.1004-6801.2025.05.019
文章导航
收藏切换

为了提高小样本图像条件下列车轮对轴承故障检测水平,提出了一种基于多分辨率孪生神经网络(multi⁃resolution siamese neural network,简称MrSNN)模型的列车轮对轴承表面缺陷机器视觉检测方法。首先,采用孪生神经网络(siamese neural network,简称SNN)为基础模型框架,构建了包含不同卷积核尺寸及不同膨胀因子大小的多分辨率卷积融合模块(multi‑resolution convolution fusion block,简称MrCFB)来综合提取图像中的细节特征与轮廓特征信息;其次,采用通道和空间的双重注意力机制重新标定多分辨率特征权重,进一步增强模型的图像特征提取能力;最后,通过对轮对轴承正常、划伤、凹痕及剥落4类图像进行检测分析,完成了算法验证。实验结果表明,轮对轴承3类故障图像识别率为100%,正常图像识别率为95%,总体识别准确率为98.75%,识别准确率优于传统SNN和YOLO‑V5等网络模型。

滚动轴承  /  故障诊断  /  神经网络  /  图像检测  /  多分辨率特征

In order to improve the fault detection performance of wheelset bearings under small sample image conditions,a machine vision inspection method based on a novel multi-resolution siamese neural network (MrSNN) is proposed for surface defect detection of wheelset bearings. First,the siamese neural network (SNN) is used as the basic model framework. A multi-resolution convolution fusion block (MrCFB) containing convolution kernels of different sizes and dilation factors is constructed to comprehensively extract the detailed features and contour features from images. Then,a dual attention mechanism combining channel and spatial information is adopted to recalibrate the multi-resolution feature weights,further enhancing the image feature extraction capability of the model. Finally,the algorithm is validated through the detection and analysis of four types of wheelset bearings images: normal,scratched,pitted and spalled. Experimental results show that the recognition rate for the three types of faulty images reaches 100%,the recognition rate for normal images is 95%,and the overall recognition accuracy is 98.75%. The recognition accuracy is superior to that of traditional SNN and YOLO-V5 models.

rolling bearing  /  fault diagnosis  /  neural network  /  image detection  /  multi-resolution feature
邓飞跃, 毕妍, 刘永强, 宋春雪, 郝如江. 基于多分辨率孪生网络的轮对轴承故障检测. 振动、测试与诊断, 2025 , 45 (5) : 1001 -1007 . DOI: 10.16450/j.cnki.issn.1004-6801.2025.05.019
Feiyue DENG, Yan BI, Yongqiang LIU, Chunxue SONG, Rujiang HAO. Wheelset Bearing Fault Detection Based on Multi‑resolution Siamese Network[J]. Journal of Vibration,Measurement and Diagnosis, 2025 , 45 (5) : 1001 -1007 . DOI: 10.16450/j.cnki.issn.1004-6801.2025.05.019
轮对轴承是列车走行部的核心部件之一,其健康状态直接影响列车运行的安全水平。然而,恶劣的工作环境极易引发列车轮对轴承发生划伤、剥离及凹痕等多种故障1。因此,为了保障列车的安全平稳运行,亟待对轮对轴承的各类缺陷故障进行准确检测。当前,振动信号分析2、轴箱温度监测3及轨边声学监测4等技术被广泛用于列车轴承的故障诊断。但是,受安装空间狭小、缺少电源等硬件条件的限制,铁路列车上安装车载振动监测系统非常困难。轴箱温度监测存在频发的误报警现象,并且无法有效监测轮对轴承早期故障。轮对轴承布置在轴箱内部,不作解体检查时,很难发现轴承表面的损伤。因此,列车必须按照运行里程及时间有计划地返厂进行拆解维修,才能更准确地检测轴承故障。现阶段,主要按照《滚动轴承圆锥滚子技术条件》,通过人工视觉及经验检测轴承表面缺陷故障。
基于机器视觉的表面缺陷检测方法具有速度快、成本低及智能化等特点,能够适应现代化工业生产,正逐步取代人工检测方法。王恒迪等5对采集的列车轴承表面缺陷图像进行中值滤波、阈值分割及边缘检测处理后,利用正常与缺陷图片的灰度差值来识别缺陷轴承。陈硕等6采用四连通域确定相关区域后,通过最小二乘法识别列车轴承套圈轮廓并进行磕碰伤检测。石炜等7对列车轴承图像进行二值化处理、形态学滤波及图像标记后,利用分类决策树进行不同类型缺陷识别。杨加东等8对图像进行阈值分割与几何特征提取后,构建反向传播(back propagation,简称BP)神经网络对列车轴承缺陷进行分类识别。上述检测方法需要繁杂的图像预处理操作,无法快速地对图像进行分析处理。此外,传统机器学习方法存在图像特征提取困难、学习能力差等问题,不能满足列车轮对轴承表面缺陷检测精度、速度及泛化性的实际需求。
随着深度学习理论的发展,基于深度神经网络开展图像识别的研究已经成为机器视觉领域中的热点。Xie等9构建了两级分层的深度卷积神经网络(convolutional neural network,简称CNN)用于下水管道缺陷的识别与分类。Wang等10提出了一种双向联合CNN检测模型,能够较为有效地提取不同损伤图片的纹理特征,识别钢板表面缺陷的不同类型。此外,YOLO系列网络11、Transformer模型12等方法也不断涌现。在实际场景中,轮对轴承表面缺陷故障复杂多样,难以获得一定规模的样本图像,传统的深度学习方法在这种小样本任务学习中的检测准确率往往不能令人满意。
基于上述分析,笔者提出了一种特征提取增强的MrSNN模型构建方法,用于解决小样本图像下轮对轴承表面缺陷难以准确检测的问题。该模型的主要优势在于:①通过搭建SNN双分支网络框架,以相似性度量为判别依据,在小样本条件下对未知图像样本进行有效识别,克服了传统神经网络模型在小样本数据集下准确率不高的问题;②构建包含不同卷积核尺寸及不同膨胀因子大小的MrCFB,通过增加网络模型的宽度,综合提取图像中的细节特征与轮廓特征,从而获取鲁棒性更好的多分辨率特征信息,增强特征识别效果;③建立了基于通道和空间的双重注意力机制,可以自适应地对已提取的多分辨率特征进行标定,在增加重要特征权重占比的同时,抑制无效特征及图像噪声的干扰,从两个维度上增强了特征选取的准确性。
传统的CNN已被证明具有较强的特征挖掘与特征学习能力,广泛用于文本分类、模式识别等领域13。然而,传统的CNN往往依赖大量的数据样本以及样本标签,只有在大样本数据驱动下才能得到学习能力较好的网络训练模型。轮对轴承检修过程中出现缺陷的概率较小,具有很强的随机性,大部分都是正常轴承,难以获取一定规模的具有明确标签的样本图像。因此,要准确识别轮对轴承表面缺陷故障,需要构建适用于小样本条件下的图像识别神经网络模型。
SNN是一种基于相似性度量的网络模型,具有很强的未知样本判别能力,已经在外貌识别、签名验证及视觉跟踪等领域得到了初步应用1415。由于SNN擅于处理小样本问题,因此通过构建SNN框架来实现小样本图像的分类识别。与CNN模型不同,SNN由两个分支结构组成,以其作为特征提取的网络具有相同的网络结构,共享权值参数。孪生神经网络框架如图1所示。在进行图像任务识别过程中,一个分支输入参考的样本图像,另一个分支输入待测试的样本图像,这两个图像构成了一个样本对。两路特征提取网络分别提取的特征信息,映射输出至同一低维特征空间,分别得到相应的特征向量。计算这一特征向量对之间的欧氏距离,并以此判别样本对之间的相似程度。其数学表达式为
其中:为样本对之间的欧氏距离;为特征向量之间的二范数。
当样本对中接近、同属于一个类别时,其相似程度大;当分属于不同类别时,其相似程度小。对于数量为N的训练样本,通过SNN框架可以构建总数为N×(N-1)个样本对,有效增大了用于训练的样本数量,提升了小样本条件下SNN训练结果的鲁棒性。
SNN模型在训练过程中采用对比损失函数作为目标优化函数,其表达式为
其中:m为设定阈值;Y为两样本之间相似度度量标签。
Y=1表示中两个样本相似,较大,说明当前网络模型不好,损失加大;Y=0表示两个图像样本不相似,较小,损失亦会增大16
传统的SNN模型中特征提取网络通常采用AlexNet。小样本条件下要准确识别轮对轴承表面缺陷的各类故障,需要充分挖掘有限数量图像中的特征信息。传统的CNN中卷积层往往采用单一尺度的卷积操作,卷积过程中感受野的范围是固定的,只能提取单一分辨率的图像特征信息。同时,固定尺度的卷积操作只能在单通道内进行,因此限制了网络卷积层的宽度。
增大卷积操作中感受野的范围可以通过增大卷积核尺寸及改变膨胀因子大小来实现。标准卷积(standard convolution,简称SC)中卷积核的尺寸与感受野的范围相同,增大卷积核尺寸,卷积操作中感受野的范围也会相应增大。不同卷积核尺寸的卷积操作如图2所示。增大膨胀因子则是增大了卷积操作中特征图各元素之间的间隔,不同膨胀因子的卷积操作如图3所示。虽然图中卷积核的尺寸都是s=2×2,但随着膨胀因子d的增大,感受野的范围显著增加,这被称为膨胀卷积(dilation convolution,简称DC)。对比发现,通过改变膨胀因子来增大卷积感受野的效率明显高于增大卷积核尺寸,但是膨胀因子越大,卷积操作中遗漏的元素数量也越多,虽然整个轮廓增大了,但细节特征也会明显丢失。
笔者提出一种新的MrCFB构建方法,该模块融合了不同膨胀因子大小与不同卷积核尺寸的卷积操作。多分辨率卷积融合模块如图4所示。整个模块采用残差块结构,通过添加恒等映射有效缓解了网络深度增加带来的梯度消失及网络性能退化问题。首先,在主连接中同一层采用了卷积核尺寸为s=3×3的4个DC操作,膨胀因子d分别为1、2、3和4,随着膨胀因子不断增大,相应感受野范围明显增大,所提取的图像轮廓范围也会增大;其次,利用Concatenate级联操作将不同分辨率的图像轮廓特征融合在一起;然后,进行4个卷积核尺寸分别为1×1、3×3、5×5和7×7的SC操作,提取图像不同分辨率的细节特征信息;最后,再次通过级联操作将多分辨率特征信息进行融合。为进一步提升网络的泛化能力,加快收敛速度,卷积操作之前采用了批标准(batch normalization,简称BN)操作和函数为ReLU的激活操作。所提模块有效增加了网络的宽度,通过融合不同膨胀因子的DC及不同卷积核尺寸的SC操作,综合提取了图像中的轮廓特征和细节特征信息。
多分辨卷积融合模块虽然可以充分挖掘图像中的轮廓特征和细节特征信息,但是由于级联融合操作导致提取到的图像特征信息成倍增加。因此,突出有用特征的权重,减少无用特征的占比,将提高关键特征提取的准确率,增强网络模型的性能。双重注意力机制是一种用于输入特征增强的方法17,相比于常用的压缩激活(squeeze and excitation,简称SE)注意力机制,其同时引入了通道注意力模块和空间注意力模块,均采用了全局最大池化(global max pooling,简称GMP)和全局平均池化(global average pooling,简称GAP)操作。相比于采用单一的GMP或GAP,其具有更强的特征表征能力18
双重注意力机制模块如图5所示,由通道和空间两个注意力模块顺序连接。通道注意力模块中,首先对特征图分别进行GMP和GAP操作,将得到的特征图分别送入一个2层的共享参数的多层感知(multilayer perceptron,简称MLP)网络,然后将输出的结果相加并输入至Sigmoid函数,得到归一化后权重系数。具体过程表示为
其中:为Sigmoid归一化操作。
将权重系数与原特征图在对应通道相乘得到处理后的特征图,并将其作为空间注意力机制模块输入特征图。基于通道分别进行GMP和GAP操作,得到两个特征图,通过级联操作堆叠在一起,再经过卷积核数量为1、尺寸为s=7×7的SC操作后,通过Sigmoid函数处理得到权重系数,即
其中:为7×7卷积。
双重注意力机制从通道和空间两个维度自适应地对特征图不同位置赋予了不同的权重系数,提高了特征图中各个特征在通道和空间上的联系。本研究将其嵌入MrCFB中,筛选出更有价值的关键特征信息,有效增强了整个模块的特征提取能力。
实验中轮对轴承来自于中车石家庄车辆有限公司轮对轴承检修流水线,轴承样本图像包含正常、划伤、凹痕及剥落4种类型。部分轮对轴承图像如图6所示。由于列车轮对轴承发生故障具有随机性强、概率性小等特点,因此所采集的缺陷图像样本数量较少,4种类别图像共80张。为提升模型分析的有效性,通过数据增强的方式对图像进行扩充。首先,分别采用水平翻转和垂直翻转方式使原有全部图像数量扩充了2倍;其次,对原有图像进行平移,通过采用调整图像色度和对比度的方式使平移后的图像再扩充2倍,最终得到4类样本图像400张。实验分析时,每一类别中随机选取训练集样本占80%,验证集样本占10%,测试集样本占10%。实验分析采用五折交叉验证,最终结果取均值。
本研究基于SNN框架,所构建的基于特征增强的特征提取网络如图7所示。网络首层是卷积核个数为32、尺寸为3×3、步长为4的SC操作,经过BN和ReLU激活操作后,连接2个特征增强的MrCFB。第1个模块中每个分支卷积核个数为16,融合后共64通道;第2个模块中每个分支卷积核个数为32,共128个。最后是2个全连接层(fully connected layer,简称FC)。第1个FC神经元的个数为128,第2个FC神经元个数为50,激活函数为Sigmoid。为了减少过拟合的影响,FC前均使用了Dropout层,参数为0.5。模型分析使用的是Python3.6编程语言,深度学习框架为tensorflow。计算机硬件配置为Intel(R) Core i7‑7700 CPU,8GB内存。模型训练时,数据批量大小为64,轮次为150,学习率参数为0.000 1。
传统孪生网络仅具备二分类性能,即区分测试样本对是否属于同一类。如果测试样本对属于同一类,测试标签为1,是正例识别;不属于同一类,测试标签为0,是负例识别。评价孪生网络二分类结果准确率P的计算式为
其中:TP为将正例识别为正例的数量;FP为将负例错误识别为正例的数量。
为了使所提网络模型具备多分类的能力,在所提网络模型二分类的基础上进行如下操作:
1) 在训练集中随机构建样本对,两图像属于同一类,标签为1,否则标签为0;
2) 遍历测试集中图像,将其作为待测样本图像,依次从测试集4类图像样本中随机选取1张图像,与待测样本构成测试样本对;
3) 搭建所提网络模型,输入训练样本及标签,完成网络的迭代训练,得到训练好的网络模型;
4) 分别记录测试样本对中待测图像样本与非待测图像样本的类别,然后将测试样本对输入已训练好的网络模型进行测试;
5) 根据网络模型测试输出,判断测试样本对的检测结果,计算每一类待测图像的识别准确率。
所提网络模型在训练过程中,其损失值迭代过程如图8所示。由图可知,损失值(无量纲)随着轮次的增加而减小,在轮次为100之后,训练集和验证集的损失值趋于稳定,网络模型达到收敛。
通过所提网络模型对测试集样本进行识别,缺陷识别准确率见表1。测试结果混淆矩阵如图9所示。可知,轮对轴承划伤、凹痕及剥落故障中每一类的识别准确率均可达到100%,正常样本的识别准确率为95%,所有测试样本整体识别准确率为98.75%。通过对比分析发现,部分正常样本图像被错误划分为划伤缺陷,这是因为列车轮对轴承长期服役过程中,轴承元件表面出现了类似划伤的痕迹,但由于程度轻微,工业现场中并不会判别为划伤缺陷。
批量大小的设置会对所提模型的检测准确率和运行效率产生影响。将其分别设置为16、32、64、128和256,不同批量大小的影响结果如图10所示。由图可知,批量大小设置的越大,模型的训练时间越少,当设置为64时,检测准确率最高,因此本研究取批量大小为64。网络模型中MrCFB数量越多,网络深度越深,结构越复杂,运行效率越低;反之,深度越浅,挖掘特征的能力越弱。因此,有必要分析MrCFB模块数量对所提网络模型的影响。模块数量分别设置为1,2,3和4,不同模块数量的影响结果如图11所示。由图可知,随着模块数量的增加,模型训练所需的时间也明显增加,运行效率显著降低。当模块数量为2,3,4时,检测准确率都接近100%,说明模型的识别效果令人满意。因此,本研究在综合考虑网络模型运行效率和准确率的前提下,MrCFB数量设置为2。
所提网络模型中阈值m(无量纲)的设定也会对模型输出结果有较大影响,不同阈值的影响结果如图12所示。由图可知:当m值设定为0.6及以下时,模型输出准确率较低;当m值设定为0.7及以上时,模型输出准确率较高,当m=0.8时准确率最高,因此本研究选取m=0.8。
对目前常用的4种方法进行了对比分析,即分别以Alexnet19和ResNet20为主干网络的SNN模型,YOLO‑V521网络模型及主干网络为所提MrCFB(不包含双重注意力机制)的SNN模型。识别准确率结果对比如图13所示。由图可知,本研究所提方法对轮对轴承表面缺陷的检测准确率最高。Alexnet没有采用残差结构,而且每层卷积中卷积核大小不变,因此在特征提取能力及网络训练梯度消失方面均弱于其他模型;ResNet中每层卷积核大小也是固定的,仅能提取特定尺度下的图像特征信息,针对内容较为复杂的轴承损伤图像,特征信息提取能力存在明显不足;YOLO‑V5网络模型虽然具有很强的图像特征检测与提取能力,但需要大规模的数据才能得到较好的网络模型训练结果,并不擅长分析小样本图像;本研究方法通过MrCFB融合了不同膨胀因子大小与不同卷积核尺寸的卷积操作,可以有效提取复杂图像中的轮廓特征和细节特征信息,并且结合双重注意力机制,进一步提升了模型挖掘特征信息的质量及特征提取能力,因此面对复杂的轮对轴承表面缺陷图像识别,本研究所提模型能够取得较高的识别准确率。
1) 提出了特征提取增强的多分辨率MrSNN模型构建方法,其基于MrCFB与双重注意力机制模块,在小样本条件下具有较强的未知图像样本识别能力。
2) 所构建的MrCFB融合了不同膨胀因子大小与不同卷积核尺寸的卷积操作,能够综合提取图像中的细节特征与轮廓特征信息,并且嵌入了由通道注意力和空间注意力组成的双重注意力机制模块,进一步增强了图像特征提取能力。
3) 通过对工业现场采集的列车轮对轴承小样本图像进行故障检测分析,验证了所提方法的有效性。结果表明,所提方法故障检测准确率优于传统SNN、YOLO‑V5等网络模型。
  • 国家自然科学基金资助项目(12272243)
参考文献 引证文献
排序方式:
1
HUANG W TSUN H JLUO J Net al. Periodic feature oriented adapted dictionary free OMP for rolling element bearing incipient fault diagnosis[J]. Mechanical Systems and Signal Processing2019126: 137-160.
2
李永健, 宋浩, 李鹏, . 改进离散熵在列车轴承损伤检测中的应用[J]. 振动、测试与诊断202343(2): 304-311.
LI YongjianSONG HaoLI Penget al. Application of improved dispersion entropy to fault detection of axle‑box bearing in train[J]. Journal of Vibration, Measurement & Diagnosis202343(2): 304-311.(in Chinese)
3
YAN GYU C QBAI Y. A new hybrid ensemble deep learning model for train axle temperature short term forecasting[J]. Machines20219(12): 312.
4
ZHANG S BHE Q BZHANG H Bet al. Doppler correction using short-time MUSIC and angle interpolation resampling for wayside acoustic defective bearing diagnosis[J]. IEEE Transactions on Instrumentation and Measurement201766(4): 671-680.
5
王恒迪, 李莎, 邓四二, . 轴承外圈侧面缺陷的视觉检测算法研究[J]. 机械设计与制造2017(12): 169-172.
WANG HengdiLI ShaDENG Sieret al. Research on visual inspection algorithm of bearing outer ring side defects[J]. Machinery Design & Manufacture2017(12): 169-172.(in Chinese)
6
陈硕, 林志敏, 吴岳彬, . 轴承套圈端面缺陷在线视觉检测的研究与实现[J]. 轴承2022(2): 48-54.
CHEN ShuoLIN ZhiminWU Yuebinet al. Research and implementation of on-line visual inspection of defects on end face of bearing rings[J]. Bearing2022(2): 48-54.(in Chinese)
7
石炜, 张袁祥, 李嘉楠. 列车滚子轴承表面缺陷机器视觉检测方法研究[J]. 机械设计与制造2022374(4): 183-186.
SHI WeiZHANG YuanxiangLI Jianan. Research on machine vision detection method for surface defects of train roller bearings[J]. Machinery Design & Manufacture2022374(4): 183-186.(in Chinese)
8
杨加东, 谢明, 王丽华, . 基于BP神经网络的表面缺陷检测分类[J]. 机床与液压201745(16): 160-164.
YANG JiadongXIE MingWANG Lihuaet al. Surface defect detection and classification based on BP neural network[J]. Machine Tool & Hydraulics201745(16): 160-164.(in Chinese)
9
XIE QLI D WXU J Xet al. Automatic detection and classification of sewer defects via hierarchical deep learning[J]. IEEE Transactions on Automation Science and Engineering201916(4): 1836-1847.
10
WANG TCHEN YQIAO M Net al. A fast and robust convolutional neural network-based defect detection model in product quality control[J]. The International Journal of Advanced Manufacturing Technology201894(9): 3465-3471.
11
TIAN Y NYANG G DWANG Zet al. Apple detection during different growth stages in orchards using the improved YOLO-V3 model[J]. Computers and Electronics in Agriculture2019157: 417-426.
12
YANG G LTANG HDING M Let al. Transformer-based attention networks for continuous pixel-wise prediction[C]∥2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada: IEEE, 2021: 16249-16259.
13
JIA FLEI Y GLU Net al. Deep normalized convolutional neural network for imbalanced fault classification of machinery and its understanding via visualization[J]. Mechanical Systems and Signal Processing2018110: 349-367.
14
ZHANG CLIU WMA H Det al. Siamese neural network based gait recognition for human identification[C]∥2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Shanghai, China: IEEE, 2016: 2832-2836.
15
LI M DCHANG KBEARCE Bet al. Siamese neural networks for continuous disease severity evaluation and change detection in medical imaging[J]. NPJ Digital Medicine20203: 48.
16
FU R RWANG Y DJIA C C. Data augmentation for cross-subject EEG features using Siamese neural network[J]. Biomedical Signal Processing and Control202275: 103614.
17
LI, ZCHEN, HLIN, Ket al. From edge data to recommendation: A double attention-based deformable convolutional network[J]. Peer-to-Peer Networking and Applications202114(6): 3984-3997.
18
CHEN L JYAO H DFU J Yet al. The classification and localization of crack using lightweight convolutional neural network with CBAM[J]. Engineering Structures2023275: 115291.
19
LU S YLU Z HZHANG Y D. Pathological brain detection based on AlexNet and transfer learning[J]. Journal of Computational Science201930: 41-47.
20
HE K MZHANG X YREN S Qet al. Deep residual learning for image recognition[C]∥2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 770-778.
21
FANG Y MGUO X XCHEN Ket al. Accurate and automated detection of surface knots on sawn timbers using YOLO-V5 model[J]. BioResources202116(3): 5390-5406.
2025年第45卷第5期
PDF下载
84
40
引用本文
BibTeX
文章信息
doi: 10.16450/j.cnki.issn.1004-6801.2025.05.019
  • 接收时间:2023-03-20
  • 首发时间:2026-03-27
  • 出版时间:2025-10-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2023-03-20
  • 修回日期:2023-08-26
基金
国家自然科学基金资助项目(12272243)
作者信息
    1石家庄铁道大学机械工程学院 石家庄,050043
    2中车石家庄车辆有限公司 石家庄,050043
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/zdcsyzd/CN/10.16450/j.cnki.issn.1004-6801.2025.05.019
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏