Article(id=1154429173130056194, tenantId=1146029695717560320, journalId=1146119893612605453, issueId=1154429165668393604, articleNumber=null, orderNo=null, doi=null, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1658332800000, receivedDateStr=2022-07-21, revisedDate=null, revisedDateStr=null, acceptedDate=null, acceptedDateStr=null, onlineDate=1753166963130, onlineDateStr=2025-07-22, pubDate=1721404800000, pubDateStr=2024-07-20, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1753166963130, onlineIssueDateStr=2025-07-22, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1753166963130, creator=13701087609, updateTime=1753166963130, updator=13701087609, issue=Issue{id=1154429165668393604, tenantId=1146029695717560320, journalId=1146119893612605453, year='2024', volume='42', issue='7', pageStart='853', pageEnd='994', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1753166961351, creator=13701087609, updateTime=1753694581412, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1156642164642796057, tenantId=1146029695717560320, journalId=1146119893612605453, issueId=1154429165668393604, language=EN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1156642164646990362, tenantId=1146029695717560320, journalId=1146119893612605453, issueId=1154429165668393604, language=CN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=915, endPage=922, ext={EN=ArticleExt(id=1154429174111523353, articleId=1154429173130056194, tenantId=1146029695717560320, journalId=1146119893612605453, language=EN, title=Early fault warning strategy for offshore wind turbine bearings based on transfer learning, columnId=null, journalTitle=Renewable Energy Resources, columnName=null, runingTitle=null, highlight=null, articleAbstract=

A transfer learningbased early fault warning method for offshore wind turbine bearings is established to address the problems of varying operating conditions of offshore wind turbines and many false alarms for early fault warning of turbine bearings. The method uses the shorttime Fourier transform to extract the timefrequency domain features of the vibration signals, which are normalised to form pre processed samples. The objective function of the convolutional autoencoder is supplemented with a support vector data description regular term and a maximum mean discrepancy regular term to constrain the feature distribution while obtaining the common features center of the bearings in normal state under different operating conditions. The Euclidean distance between the online sample features and the common feature center is calculated to construct bearing health indicator sequence, and the ADF(Augmented DickeyFuller)test is introduced to perform stationarity analysis and capture the sequence mutation points, which finally realize the early fault warning of bearings in offshore wind turbines. The validation on the XJTUSY bearing dataset showed that the proposed method has fewer false alarms, high accuracy and better detection stability.

, 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=Zhicheng Xin, Longjun Wang, Shenquan Liu), CN=ArticleExt(id=1154429202347578106, articleId=1154429173130056194, tenantId=1146029695717560320, journalId=1146119893612605453, language=CN, title=基于迁移学习的海上风电机组轴承早期故障预警策略, columnId=null, journalTitle=可再生能源, columnName=null, runingTitle=null, highlight=null, articleAbstract=

针对海上风电机组工况不一、机组轴承早期故障预警误报警多的问题,文章建立了一种基于迁移学习的海上风电机组轴承早期故障预警方法。首先,采用短时傅立叶变换提取振动信号时频域特征,归一化后形成预处理样本;然后,在卷积自编码器的目标函数中添加支持向量数据描述正则项和最大均值差异正则项,约束特征分布的同时获得轴承在不同工况正常状态的公共特征中心;最后,计算在线样本特征与公共特征中心的欧氏距离,构建轴承健康指标序列,引入增广迪基富勒检验(ADF)方法作平稳性分析,捕捉序列突变点,最终实现对海上风电机组轴承早期故障预警。在 XJTUSY 轴承数据集上的验证表明,所提方法误报警少、准确度高,具有更好的检测稳定性。

, correspAuthors=null, authorNote=null, correspAuthorsNote=
汪隆君(1982-),男,博士,讲师,研究方向为电力系统可靠性与规划。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=+hp6Up82WzM4Ah39IPWNbg==, magXml=oV/919CAWoSXgMIiI1uSsw==, pdfUrl=null, pdf=RlWlh0WeYT7pzAi9HsZvNg==, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=UR/N2LQYgMtNxtksiNitNQ==, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=辛治铖, 汪隆君, 刘沈全)}, authors=[Author(id=1154429204197266191, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, orderNo=0, 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=1154429204285346576, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204197266191, language=EN, stringName=Zhicheng Xin, firstName=Zhicheng, middleName=null, lastName=Xin, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1154429204360844049, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204197266191, language=CN, stringName=辛治铖, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=null), Author(id=1154429204406981395, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=epwlj@scut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1154429204503450388, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204406981395, language=EN, stringName=Longjun Wang, firstName=Longjun, middleName=null, lastName=Wang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1154429204566364949, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204406981395, language=CN, stringName=汪隆君, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=null), Author(id=1154429204629279511, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, 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=1154429204738331416, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204629279511, language=EN, stringName=Shenquan Liu, firstName=Shenquan, middleName=null, lastName=Liu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1154429204792857369, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, authorId=1154429204629279511, language=CN, stringName=刘沈全, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=null, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=null)], keywords=[Keyword(id=1154429205438780186, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, orderNo=1, keyword=early fault warning), Keyword(id=1154429205484917531, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, orderNo=2, keyword=stability test), Keyword(id=1154429205535249180, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, orderNo=3, keyword=transfer learning), Keyword(id=1154429205593969440, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, orderNo=4, keyword=bearing), Keyword(id=1154429205644301092, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, orderNo=5, keyword=offshore wind), Keyword(id=1154429205698827049, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, orderNo=1, keyword=早期故障预警), Keyword(id=1154429205749158697, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, orderNo=2, keyword=平稳性检验), Keyword(id=1154429205799490347, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, orderNo=3, keyword=迁移学习), Keyword(id=1154429205845627693, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, orderNo=4, keyword=轴承), Keyword(id=1154429205954679601, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, orderNo=5, keyword=海上风电)], refs=[Reference(id=1154429208303489883, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=null, pageEnd=null, url=https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2020-full-report.pdf, language=null, rfNumber=[1], rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=British Petroleum. Statistical review of world energy 2020[EB/OL]. https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2020-full-report.pdf2022-07-20., articleTitle=Statistical review of world energy 2020, refAbstract=null), Reference(id=1154429208353821532, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[2], rfOrder=1, authorNames=Global Wind Energy Council, journalName=null, refType=null, unstructuredReference=Global Wind Energy Council. Global offshore wind report 2020[R]. Brussel: Global Wind Energy Council, 2021., articleTitle=Global offshore wind report 2020, refAbstract=null), Reference(id=1154429208437707613, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2016, volume=80, issue=null, pageStart=349, pageEnd=376, url=null, language=null, rfNumber=[3], rfOrder=2, authorNames=Zhang H, Chen X, Du Z, journalName=Mechanical Systems and Signal Processing, refType=null, unstructuredReference=Zhang H, Chen X, Du Z, et al. Kurtosis based weighted sparse model with convex optimization technique for bearing fault diagnosis[J]. Mechanical Systems and Signal Processing, 2016, 80: 349-376., articleTitle=Kurtosis based weighted sparse model with convex optimization technique for bearing fault diagnosis, refAbstract=null), Reference(id=1154429208576119646, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2022, volume=190, issue=null, pageStart=110698, pageEnd=null, url=null, language=null, rfNumber=[4], rfOrder=3, authorNames=Sun H, Cao X, Wang C, journalName=Measurement, refType=null, unstructuredReference=Sun H, Cao X, Wang C, et al. An interpretable anti-noise network for rolling bearing fault diagnosis based on FSWT[J]. Measurement, 2022, 190: 110698., articleTitle=An interpretable anti-noise network for rolling bearing fault diagnosis based on FSWT, refAbstract=null), Reference(id=1154429208676782943, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2017, volume=82, issue=null, pageStart=103, pageEnd=116, url=null, language=null, rfNumber=[5], rfOrder=4, authorNames=Imaouchen Y, Kedadouche M, Alkama R, journalName=Mechanical Systems and Signal Processing, refType=null, unstructuredReference=Imaouchen Y, Kedadouche M, Alkama R, et al. A frequency-weighted energy operator and complementary ensemble empirical mode decomposition for bearing fault detection[J]. Mechanical Systems and Signal Processing, 2017, 82: 103-116., articleTitle=A frequency-weighted energy operator and complementary ensemble empirical mode decomposition for bearing fault detection, refAbstract=null), Reference(id=1154429208764863328, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=2, pageStart=68, pageEnd=74, url=null, language=null, rfNumber=[6], rfOrder=5, authorNames=汤婷, 张岩, 安宗文, journalName=轴承, refType=null, unstructuredReference=汤婷, 张岩, 安宗文. 基于孤立森林算法的风电齿轮箱轴承故障检测方法[J]. 轴承, 2022(2): 68-74., articleTitle=基于孤立森林算法的风电齿轮箱轴承故障检测方法, refAbstract=null), Reference(id=1154429208831972193, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2019, volume=32, issue=6, pageStart=102, pageEnd=104, url=null, language=null, rfNumber=[7], rfOrder=6, authorNames=翟怡萌, 李晶, 程强, journalName=工业控制计算机, refType=null, unstructuredReference=翟怡萌, 李晶, 程强, 等. 基于OCSVM的火电厂一次风机异常检测[J]. 工业控制计算机, 2019, 32(6): 102-104., articleTitle=基于OCSVM的火电厂一次风机异常检测, refAbstract=null), Reference(id=1154429208882303842, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2022, volume=48, issue=1, pageStart=302, pageEnd=314, url=null, language=null, rfNumber=[8], rfOrder=7, authorNames=毛文涛, 田思雨, 窦智, journalName=自动化学报, refType=null, unstructuredReference=毛文涛, 田思雨, 窦智, 等. 一种基于深度迁移学习的滚动轴承早期故障在线检测方法[J]. 自动化学报, 2022, 48(1): 302-314., articleTitle=一种基于深度迁移学习的滚动轴承早期故障在线检测方法, refAbstract=null), Reference(id=1154429208961995619, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2022, volume=33, issue=10, pageStart=1234, pageEnd=1243, url=null, language=null, rfNumber=[9], rfOrder=8, authorNames=赵志宏, 李乐豪, 杨绍普, journalName=中国机械工程, refType=null, unstructuredReference=赵志宏, 李乐豪, 杨绍普, 等. 一种无监督的轴承健康指标及早期故障检测方法[J]. 中国机械工程, 2022, 33(10): 1234-1243., articleTitle=一种无监督的轴承健康指标及早期故障检测方法, refAbstract=null), Reference(id=1154429209016521573, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2021, volume=148, issue=null, pageStart=1, pageEnd=6, url=null, language=null, rfNumber=[10], rfOrder=9, authorNames=Zhang Z, Deng X, journalName=Pattern Recognition Letters, refType=null, unstructuredReference=Zhang Z, Deng X. Anomaly detection using improved deep SVDD model with data structure preservation[J]. Pattern Recognition Letters, 2021, 148: 1-6., articleTitle=Anomaly detection using improved deep SVDD model with data structure preservation, refAbstract=null), Reference(id=1154429209079436135, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2021, volume=40, issue=6, pageStart=25, pageEnd=33, url=null, language=null, rfNumber=[11], rfOrder=10, authorNames=何桂雄, 金璐, 李克成, journalName=电力工程技术, refType=null, unstructuredReference=何桂雄, 金璐, 李克成, 等. 基于改进 DaNN 的综合能源系统多能负荷预测[J]. 电力工程技术, 2021, 40(6): 25-33., articleTitle=基于改进 DaNN 的综合能源系统多能负荷预测, refAbstract=null), Reference(id=1154429209175905129, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, doi=null, pmid=null, pmcid=null, year=2019, volume=55, issue=16, pageStart=1, pageEnd=6, url=null, language=null, rfNumber=[12], rfOrder=11, authorNames=雷亚国, 韩天宇, 王彪, journalName=机械工程学报, refType=null, unstructuredReference=雷亚国, 韩天宇, 王彪, 等. XJTU-SY 滚动轴承加速寿命试验数据集解读[J]. 机械工程学报, 2019, 55(16): 1-6., articleTitle=XJTU-SY 滚动轴承加速寿命试验数据集解读, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1154429204084019979, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, xref=1, ext=[AuthorCompanyExt(id=1154429204092408588, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, companyId=1154429204084019979, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 South China University of Technology Guangzhou 510000 China), AuthorCompanyExt(id=1154429204100797197, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, companyId=1154429204084019979, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 华南理工大学 广东 广州 510000)])], figs=[ArticleFig(id=1154429207242330950, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Fig. 1, caption=Time series of bearing health indicators, figureFileSmall=J5APiWLPbew+bZyr0naq+w==, figureFileBig=7uUEgRrf9GNKa0MImOGxMA==, tableContent=null), ArticleFig(id=1154429207288468296, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=图 1, caption=轴承健康指标时间序列, figureFileSmall=J5APiWLPbew+bZyr0naq+w==, figureFileBig=7uUEgRrf9GNKa0MImOGxMA==, tableContent=null), ArticleFig(id=1154429207342994250, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Fig. 2, caption=Health indicators for the life cycle of test bearings, figureFileSmall=wNvkY+6lNsOukTkvhPN8RQ==, figureFileBig=ocuXKatnipIdz4/NTfksgQ==, tableContent=null), ArticleFig(id=1154429207393325900, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=图 2, caption=轴承的全寿命周期健康指标, figureFileSmall=wNvkY+6lNsOukTkvhPN8RQ==, figureFileBig=ocuXKatnipIdz4/NTfksgQ==, tableContent=null), ArticleFig(id=1154429207452046158, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Fig. 3, caption=Bearing 1_1 full life cycle vibration signal and envelope spectrum at the moment of increasing amplitude, figureFileSmall=XmY3JwmkFbqeiuQEG+6htQ==, figureFileBig=tT6K6s+Rn/sks0OI3BxEaA==, tableContent=null), ArticleFig(id=1154429207498183504, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=图 3, caption=轴承 1_1 全寿命周期振动信号及幅值增大时刻的包络谱, figureFileSmall=XmY3JwmkFbqeiuQEG+6htQ==, figureFileBig=tT6K6s+Rn/sks0OI3BxEaA==, tableContent=null), ArticleFig(id=1154429207586263889, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Fig. 4, caption=Bearing 2_2 full life cycle vibration signal and envelope spectrum at the moment of increasing amplitude, figureFileSmall=0a6dRDCoXjS+u5zcxVXHDQ==, figureFileBig=UC68oOEF4o6RDu6E+5BS+A==, tableContent=null), ArticleFig(id=1154429207644984146, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=图 4, caption=轴承 2_2 全寿命周期振动信号及幅值增大时刻的包络谱, figureFileSmall=0a6dRDCoXjS+u5zcxVXHDQ==, figureFileBig=UC68oOEF4o6RDu6E+5BS+A==, tableContent=null), ArticleFig(id=1154429207691121491, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Fig. 5, caption=Test results for three methods, figureFileSmall=VQ9Q7l7lLqZIHD5V7veiIQ==, figureFileBig=ElN61vhow1sAE70hyIGaZA==, tableContent=null), ArticleFig(id=1154429207745647444, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=图 5, caption=3 种方法的检测结果, figureFileSmall=VQ9Q7l7lLqZIHD5V7veiIQ==, figureFileBig=ElN61vhow1sAE70hyIGaZA==, tableContent=null), ArticleFig(id=1154429207833727829, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Table 1, caption=Network parameters of convolutional autoencoder, figureFileSmall=null, figureFileBig=null, tableContent=
网络层名 层参数
L1 Conv2d $\left( {5 \times 5 \times 8}\right)$
L2 Maxpool $\left( {2 \times 2}\right)$
L3 Conv2d $\left( {5 \times 5 \times 4}\right)$
L4 Maxpool $\left( {2 \times 2}\right)$
L5 Dense(32)
L6 Upsampling(2×2)
L7 Conv2dTranspose $\left( {5 \times 5 \times 4}\right)$
L8 Upsampling(2x2)
L9 Conv2dTranspose $\left( {5 \times 5 \times 8}\right)$
L10 Upsampling(2×2)
L11 Conv2dTranspose $\left( {5 \times 5 \times 1}\right)$
), ArticleFig(id=1154429207909225302, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=表 1, caption=卷积自编码器网络参数, figureFileSmall=null, figureFileBig=null, tableContent=
网络层名 层参数
L1 Conv2d $\left( {5 \times 5 \times 8}\right)$
L2 Maxpool $\left( {2 \times 2}\right)$
L3 Conv2d $\left( {5 \times 5 \times 4}\right)$
L4 Maxpool $\left( {2 \times 2}\right)$
L5 Dense(32)
L6 Upsampling(2×2)
L7 Conv2dTranspose $\left( {5 \times 5 \times 4}\right)$
L8 Upsampling(2x2)
L9 Conv2dTranspose $\left( {5 \times 5 \times 8}\right)$
L10 Upsampling(2×2)
L11 Conv2dTranspose $\left( {5 \times 5 \times 1}\right)$
), ArticleFig(id=1154429207976334167, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Table 2, caption=Test results, figureFileSmall=null, figureFileBig=null, tableContent=
模型/ 周期 迁移学习 CAE+SVDD+ MMD 参照实验 1 CAE+SVDD 参照实验 2 CAE+SVDD 参照实验 3 CAE+SVDD+ MMD
TN TP TN TP TN TP TN TP
20 8 50 0 50 0 50 4 50
40 9 50 0 50 0 50 0 50
60 16 50 2 50 0 50 1 50
80 25 50 4 50 5 50 8 50
100 36 50 7 50 9 50 18 50
120 42 49 10 50 14 50 28 50
140 42 49 19 50 24 50 36 50
160 45 49 25 50 31 50 42 50
180 48 49 30 50 38 50 44 50
200 48 49 36 50 45 50 45 50
), ArticleFig(id=1154429208097968984, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=表 2, caption=实验结果, figureFileSmall=null, figureFileBig=null, tableContent=
模型/ 周期 迁移学习 CAE+SVDD+ MMD 参照实验 1 CAE+SVDD 参照实验 2 CAE+SVDD 参照实验 3 CAE+SVDD+ MMD
TN TP TN TP TN TP TN TP
20 8 50 0 50 0 50 4 50
40 9 50 0 50 0 50 0 50
60 16 50 2 50 0 50 1 50
80 25 50 4 50 5 50 8 50
100 36 50 7 50 9 50 18 50
120 42 49 10 50 14 50 28 50
140 42 49 19 50 24 50 36 50
160 45 49 25 50 31 50 42 50
180 48 49 30 50 38 50 44 50
200 48 49 36 50 45 50 45 50
), ArticleFig(id=1154429208156689241, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=EN, label=Table 3, caption=Quantity of false alarms for different methods, figureFileSmall=null, figureFileBig=null, tableContent=
检测方法 轴承 1_1 轴承 2_1
本文方法 0 0
方法一 103 373
孤立森林 121 96
一类支持向量机 67 170
), ArticleFig(id=1154429208227992410, tenantId=1146029695717560320, journalId=1146119893612605453, articleId=1154429173130056194, language=CN, label=表 3, caption=不同方法的误报警数, figureFileSmall=null, figureFileBig=null, tableContent=
检测方法 轴承 1_1 轴承 2_1
本文方法 0 0
方法一 103 373
孤立森林 121 96
一类支持向量机 67 170
)], attaches=null, journal=Journal(id=1146118781765529606, delFlag=0, nameCn=可再生能源, nameEn=Renewable Energy Resources, nameHistory1=null, nameHistory2=null, issn=1671-5292, eissn=, cn=21-1469/TK, coden=null, periodic=0, 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=JwZjzvOuLF6gg7XdYa5wzg==, journalPrice=null, startedYear=null, abbrevIsoEn=Renew Energy Res, journalRemark=null, publicationField=null, createdTime=null, updatedTime=1755588886026, createdBy=null, updatedBy=15831073675, firstLetterCn=R, firstLetterEn=R, subjectCode=Engineering, subjectName=工程, subjectCodeEn=Engineering, subjectNameEn=null, picCn=JwZjzvOuLF6gg7XdYa5wzg==, picEn=JbD+CPZ3qeyTMjY1NLqExg==, jcr=null, cjcr=null, exts=[JournalExt(id=1164587454112678277, language=CN, name=可再生能源, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://kzsny.com/, createdTime=1755588886049, updatedTime=1755588886049, createdBy=15831073675, updatedBy=15831073675, submissionGuidelinesUrl=https://ncny.cbpt.cnki.net/EditorEN/PromptPageInfo.aspx?t=v&c=1, submissionAuthorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=1, submissionEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, submissionReviewUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=2, submissionCeEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, submissionAeEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, option={"copyright":""}), JournalExt(id=1164587454167204230, language=EN, name=Renewable Energy Resources, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1755588886062, updatedTime=1755588886062, createdBy=15831073675, updatedBy=15831073675, submissionGuidelinesUrl=, submissionAuthorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=1, submissionEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, submissionReviewUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=2, submissionCeEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, submissionAeEditorUrl=https://ncny.cbpt.cnki.net/EditorEN/index.aspx?t=3, option={"copyright":""})], databaseList=null, tenantJournalId=1146119893612605453, websiteList=[Website(id=1148243202370429360, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146119893612605453, 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/kzsny/CN, language=CN, createTime=1751692112772, createBy=18614031015, updateTime=1753519535151, updateBy=18614031015, name=《可再生能源》-中文站点, tplId=1146099689490845704, title=可再生能源, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1148619523667816493, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202370429360, code=articleTextType, value=kx, createTime=1751781834761, updateTime=1751781834761, creator=18614031015, updator=18614031015), WebsiteProps(id=1148619523642650666, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202370429360, code=banner, value=null, createTime=1751781834755, updateTime=1751781834755, creator=18614031015, updator=18614031015), WebsiteProps(id=1148619523634262057, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202370429360, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=y1rtJt0tFAqUB4W3g4tlSw==, createTime=1751781834753, updateTime=1751781834753, creator=18614031015, updator=18614031015), WebsiteProps(id=1148619523659427884, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202370429360, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1751781834759, updateTime=1751781834759, creator=18614031015, updator=18614031015), WebsiteProps(id=1148619523651039275, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202370429360, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1751781834757, updateTime=1751781834757, creator=18614031015, updator=18614031015)]), Website(id=1155908164001353748, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146119893612605453, 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/kzsny/EN, language=EN, createTime=1753519582037, createBy=18614031015, updateTime=1753519582037, updateBy=18614031015, name=《可再生能源》-英文站点, tplId=1146101810881728533, title=Renewable Energy Resources, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1155908456914739359, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155908164001353748, code=articleTextType, value=kx, createTime=1753519651869, updateTime=1753519651869, creator=18614031015, updator=18614031015), WebsiteProps(id=1155908456893767836, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155908164001353748, code=banner, value=null, createTime=1753519651864, updateTime=1753519651864, creator=18614031015, updator=18614031015), WebsiteProps(id=1155908456885379227, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155908164001353748, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=y1rtJt0tFAqUB4W3g4tlSw==, createTime=1753519651862, updateTime=1753519651862, creator=18614031015, updator=18614031015), WebsiteProps(id=1155908456906350750, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155908164001353748, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1753519651867, updateTime=1753519651867, creator=18614031015, updator=18614031015), WebsiteProps(id=1155908456902156445, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155908164001353748, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1753519651866, updateTime=1753519651866, creator=18614031015, updator=18614031015)])], journalTitle=可再生能源, weixinUrl=null, journalUrl=null, iacademicId=null, status=0, seqNo=null, journalTitleEn=Renewable Energy Resources, journalPhotoCn=JwZjzvOuLF6gg7XdYa5wzg==, journalPhotoEn=JbD+CPZ3qeyTMjY1NLqExg==, journalFirstLetter=R, 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/kzsny/CN/Y2024/V42/I7/915, detailUrlEn=https://castjournals.cast.org.cn/joweb/kzsny/EN/Y2024/V42/I7/915, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kzsny/CN/PDF/Y2024/V42/I7/915, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kzsny/EN/PDF/Y2024/V42/I7/915, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于迁移学习的海上风电机组轴承早期故障预警策略
收藏切换
PDF下载
辛治铖 , 汪隆君 , 刘沈全
可再生能源 | 2024,42(7): 915-922
收起
收藏切换
可再生能源 | 2024, 42(7): 915-922
基于迁移学习的海上风电机组轴承早期故障预警策略
全屏
辛治铖, 汪隆君 , 刘沈全
作者信息
  • 1 华南理工大学 广东 广州 510000

通讯作者:

汪隆君(1982-),男,博士,讲师,研究方向为电力系统可靠性与规划。E-mail:
Early fault warning strategy for offshore wind turbine bearings based on transfer learning
Zhicheng Xin, Longjun Wang , Shenquan Liu
Affiliations
  • 1 South China University of Technology Guangzhou 510000 China
出版时间: 2024-07-20
文章导航
收藏切换

针对海上风电机组工况不一、机组轴承早期故障预警误报警多的问题,文章建立了一种基于迁移学习的海上风电机组轴承早期故障预警方法。首先,采用短时傅立叶变换提取振动信号时频域特征,归一化后形成预处理样本;然后,在卷积自编码器的目标函数中添加支持向量数据描述正则项和最大均值差异正则项,约束特征分布的同时获得轴承在不同工况正常状态的公共特征中心;最后,计算在线样本特征与公共特征中心的欧氏距离,构建轴承健康指标序列,引入增广迪基富勒检验(ADF)方法作平稳性分析,捕捉序列突变点,最终实现对海上风电机组轴承早期故障预警。在 XJTUSY 轴承数据集上的验证表明,所提方法误报警少、准确度高,具有更好的检测稳定性。

早期故障预警  /  平稳性检验  /  迁移学习  /  轴承  /  海上风电

A transfer learningbased early fault warning method for offshore wind turbine bearings is established to address the problems of varying operating conditions of offshore wind turbines and many false alarms for early fault warning of turbine bearings. The method uses the shorttime Fourier transform to extract the timefrequency domain features of the vibration signals, which are normalised to form pre processed samples. The objective function of the convolutional autoencoder is supplemented with a support vector data description regular term and a maximum mean discrepancy regular term to constrain the feature distribution while obtaining the common features center of the bearings in normal state under different operating conditions. The Euclidean distance between the online sample features and the common feature center is calculated to construct bearing health indicator sequence, and the ADF(Augmented DickeyFuller)test is introduced to perform stationarity analysis and capture the sequence mutation points, which finally realize the early fault warning of bearings in offshore wind turbines. The validation on the XJTUSY bearing dataset showed that the proposed method has fewer false alarms, high accuracy and better detection stability.

early fault warning  /  stability test  /  transfer learning  /  bearing  /  offshore wind
辛治铖, 汪隆君, 刘沈全. 基于迁移学习的海上风电机组轴承早期故障预警策略. 可再生能源, 2024 , 42 (7) : 915 -922 .
Zhicheng Xin, Longjun Wang, Shenquan Liu. Early fault warning strategy for offshore wind turbine bearings based on transfer learning[J]. Renewable Energy Resources, 2024 , 42 (7) : 915 -922 .
2019 年,风光主导的可再生能源装机大幅增长,在一次能源新增容量中占比超过 ${40}{\%}$ [ 1 ] ,其中海上风电年均增长率超过 ${25}{\%}$ [ 2 ] 。然而,海上风电机组不仅要经受潮湿环境和盐雾侵蚀的考验, 还长期运行在复杂多变的载荷条件下,其故障率较高。另外,海上作业条件恶劣、物资运输困难,海上风电运维难度和成本均更高。
轴承作为海上风电机组的关键零件, 在动态载荷、恶劣工况等因素相互作用下,易疲劳老化造成机械性能下降, 甚至引起机组计划外停机。因此,对轴承早期故障进行预警, 将有助于充分利用海上作业窗口期制订周全有效的运维计划, 对降低机组运维成本具有重要意义。
轴承振动信号能量小, 容易被周围噪声干扰或淹没,使故障状态特征区分度不显著。针对这一问题,轴承故障特征提取多依赖人工经验的空间变换方法,如频域多点峭度 [ 3 ] 、小波变换 [ 4 ] 、经验模态分解时等。这些方法在相对的确定场景有着较好的性能, 但自适应提取特征信息能力偏弱, 场景迁移能力有限,无法适用于海上风速噪声随机、工况有差异的场景。另一方面, 基于机器学习和统计建模的异常点检测模型,如孤立森林 [ 6 ] 、一类支持向量机回等,由于单点检测使其敏感易造成误报警,需要频繁调整阈值或模型参数。
近年来, 深度神经网络在特征提取方面表现出优异的性能, 被应用于早期故障检测。文献[ 8 ] 提出采用具有域间差异正则项的自编码器提取故障特征,基于离线数据构建预警阈值实现早期故障检测。文献[ 9 ]采用自编码器提取轴承状态特征,以 Bray-Curtis 距离作为健康指标,基于指标序列的 ${3\sigma }$ 准则判断早期故障。
为提高预警稳定性, 本文提出了一种基于迁移学习的海上风电机组轴承早期故障预警方法。 首先, 采用短时傅立叶变换提取振动信号时频域特征,进一步处理后转换成灰度图样本;然后,在卷积自编码器的目标函数中添加支持向量数据描述正则项和最大均值差异正则项,约束特征分布的同时获取轴承正常状态的公共特征中心;最后, 计算在线样本特征与公共特征中心的欧氏距离, 构建轴承健康指标序列, 引入增广迪基-富勒检验(ADF)方法作平稳性分析, 捕捉序列突变点, 最终实现对海上风电机组轴承早期故障预警。
海上风电机组轴承故障样本有限且收集困难, 通常只能采集大量的正常数据和少量的有标签异常数据, 有监督多分类学习方法难以奏效。为进行轴承振动知识特征挖掘, 基于最小化重构误差的无监督学习方法可用于轴承状态特征提取。
由于机组工况差异、环境噪声干扰等原因,采集数据服从不同的边缘分布, 直接应用同一特征提取模型, 性能显著下降。为此, 需要利用某一领域(简称源域)知识,迁移至另一个相关但不相同的领域(简称目标域),仅基于少量样本即可提升特征提取模型的性能。领域自适应是迁移学习的手段之一, 它通过探索弥合分布差异的领域不变特征, 建立从源域到目标域的特征迁移, 旨在最小化源域和目标域数据间的分布差异并提高模型泛化能力。
可用海上风电机组自带的振动传感器采集轴承振动的有限样本作训练, 获得特征提取源域预训练模型; 在机组安装调试结束后现场试运行阶段, 利用现场采集同一位置的实际工况样本作为目标域输入, 对预训练模型进行微调, 迁移学习得到目标域特征提取模型。基于迁移模型对该轴承状态特征进行提取, 形成指标序列并采用滑动窗口分析序列平稳性, 通过捕捉指标突变点即可预警轴承早期故障。
出厂调试用源域样本数量多于目标域, 可被认为是正常状态样本。本文所提方法期望充分利用源域知识提取的状态特征迁移到目标域应用, 用较少的目标域训练样本实现较高的跨领域模型特征提取性能。
短时傅里叶变换 (STFT) 是一种有效反映非平稳信号频率随时间变化的分析方法。实际中通常使用离散 STFT,给定长度为 $M$ 的窗函数 $\omega \left( m\right)$ ,让其沿序列 $x\left( n\right)$ 以步长 $l$ 滑动,则在时刻 $\tau$ 的离散 STFT 为
$\operatorname{STFT}\left({\tau , k}\right)= \mathop{\sum }\limits_{{n = 0}}^{{N - 1}}x\left( n\right)\omega \left({n -\tau }\right){\mathrm{e}}^{-\mathrm{j}{2\pi nk}/N}$
式中: $k = 0,1,2\ldots , M/2$ ,表示对应的实际频率为 $k{f}_{\mathrm{s}}/M;{f}_{\mathrm{s}}$ 为采样频率; $N$ 为时间序列长度。
然而, 直接 STFT 后的样本形成的灰度图存在较多的噪点,样本相似度偏低。为此,对 STFT 后的样本提取同一频率位置在所有时刻的最大值, 代替相应位置上的频谱幅值, 再归一化转换成灰度图作为样本。上述预处理方法相当于幅度谱投影在频率轴和幅值坐标轴所构成的平面上, 并沿着时间轴复制到所有频率位置, 可降低随机噪声对样本的影响。
自编码器是一种无监督学习模型, 其通过非线性映射将输入数据投影到低维空间,并从中尽可能重构输入数据, 以达到数据降维的目的。自编码器包含编码器 ${\phi }_{\mathrm{e}}$ 和解码器 ${\phi }_{\mathrm{d}}$ 两部分,给定输入 $x$ ,自编码器的编码和解码过程分别为
$ z ={\phi }_{\mathrm{e}}\left({x;{\theta }_{\mathrm{e}}}\right)$
$\widehat{x}= {\phi }_{\mathrm{d}}\left({z;{\theta }_{\mathrm{d}}}\right)$
式中: $z$ 为隐藏空间特征; $\widehat{x}$ 为重构数据; ${\theta }_{\mathrm{e}},{\theta }_{\mathrm{d}}$ 分别为编码器和解码器参数。
由于样本采用时频域二维信息, 适合选用卷积核实现局部感知,为此,本文构建了卷积自编码, 结构如 表 1 所示。除最后一个反卷积层输出仅使用 Sigmoid 激活函数, 其余每个卷积层和反卷积层输出均使用批量归一化层和激活函数 Leaky_ReLU。
卷积自编码器的训练目标是使输入和输出的重构误差最小。
${L}_{\mathrm{{ae}}}= \frac{1}{N}\mathop{\sum }\limits_{{i = 1}}^{N}\parallel x -\widehat{x}{\parallel }_{2}^{2}+ \lambda \parallel W{\parallel }_{2}^{2}$
式中: $N$ 为样本总数; $\lambda \parallel W{\parallel }_{2}^{2}$ 为权重衰减正则项; $\lambda$ 为权重系数; $W$ 为自编码器的网络参数。
自编码器并不能保证正常样本在特征空间的紧凑性, 难以使用隐藏空间特征到公共特征中心的欧氏距离区分异常样本。为此,在自编码器的损失函数中添加支持向量数据(SVDD)描述正则项 [ 10 ] ,对特征分布的几何形状加以约束,同时由于自编码重构误差项的存在, 可避免出现特征空间坍塌现象。SVDD 正则项为
${L}_{\text{svdd }}= \frac{1}{N}\mathop{\sum }\limits_{{i = 1}}^{N}{\begin{Vmatrix}{\phi }_{\mathrm{e}}\left({x}_{i}\right)- \mathbf{c}\end{Vmatrix}}_{2}^{2}$
$ c =\frac{1}{N}\mathop{\sum }\limits_{{i = 1}}^{N}{\phi }_{\mathrm{e}}\left({x}_{i}\right)$
式中: ${\phi }_{\mathrm{e}}\left({x}_{i}\right)$ 为编码器提取的样本特征; $\mathbf{c}$ 为公共特征中心向量。
领域自适应法在应用中难以保证源域与目标域样本的均衡, 存在某些工况样本较多而某些工况样本较少的现象, 若用较少样本的工况数据直接训练模型, 可能会在该工况的特征提取过程中取得较差的性能及效果。为此, 在迁移学习模型训练的损失函数中引入最大均值差异(MMD)正则项 [ 11 ] ,通过最小化不同工况样本分布之间的距离, 诱导出不同工况的公共特征空间, 最终提高少样本工况下的识别准确率。若 2 个工况样本表示为 ${X}_{s}= {\left\{{x}_{i}^{\left( s\right)}\right\}}_{i = 1}^{{n}_{s}}$${X}_{t}= {\left\{{x}_{j}^{\left( t\right)}\right\}}_{j = 1}^{{n}_{t}}$ ,则其 MMD 可表示为
${L}_{\mathrm{{mmd}}}= {\mathrm{{MMD}}}^{2}\left({{X}_{s},{X}_{t}}\right)= \frac{1}{{n}_{s}^{2}}\mathop{\sum }\limits_{{i = 1}}^{{n}_{s}}\mathop{\sum }\limits_{{j = 1}}^{{n}_{t}}k\left({{x}_{i}^{\left( s\right)},{x}_{j}^{\left( t\right)}}\right)+ \\\frac{1}{{n}_{t}^{2}}\mathop{\sum }\limits_{{i = 1}}^{{n}_{t}}\mathop{\sum }\limits_{{j = 1}}^{{n}_{t}}k\left({{x}_{i}^{\left( s\right)},{x}_{j}^{\left( t\right)}}\right)- \frac{2}{{n}_{s}{n}_{t}}\mathop{\sum }\limits_{{i = 1}}^{{n}_{s}}\mathop{\sum }\limits_{{j = 1}}^{{n}_{t}}k\left({{x}_{i}^{\left( s\right)},{x}_{j}^{\left( t\right)}}\right)$
式中: ${n}_{s},{n}_{t}$ 分别为源域和目标域的样本数量; ${x}_{i}^{\left( s\right)}$ , ${x}_{j}^{\left( t\right)}$ 分别为源域和目标域样本; $k\left({\cdot ,\cdot }\right)$ 为核函数, 通常选择高斯核 $k\left({x,{x}^{\prime }}\right)= \exp \left({-{\begin{Vmatrix}x -{x}^{\prime }\end{Vmatrix}}^{2}/2{\sigma }^{2}}\right)$
本文建立的深度迁移卷积自编码器损失函数为
$ L ={L}_{\mathrm{{ae}}}+ \alpha {L}_{\mathrm{{svdd}}}+ \beta {L}_{\mathrm{{mmd}}}$
式中: $\alpha ,\beta$ 为不同正则项的权重系数, $\alpha > 0,\beta > 0$ ,。
选择自编码器隐藏空间特征到公共特征中心的欧氏距离平方值作为轴承的健康指标 $\mathrm{{HI}}$
$\mathrm{{HI}}= {\begin{Vmatrix}{\phi }_{\mathrm{e}}\left({x}_{i}\right)- c\end{Vmatrix}}_{2}^{2}$
式中: ${\phi }_{\mathrm{e}}, c$ 分别为迁移学习阶段获得的编码器和公共特征中心; $x$ 为在线样本。
图 1 为轴承健康指标全寿命周期变化曲线。
若采用异常点检测的方法, 则容易频繁误报警,如果将时序片段的平稳度作为异常检测指标, 可在一定程度上提升模型识别早期故障的准确性。因此,引入 ADF 检验健康指标平稳度,预警轴承早期故障。
ADF 检验假设时间序列 ${\left\{{y}_{t}\right\}}_{t =- \infty }^{\infty }$ 可由下列 3 种方程之一生成。
${y}_{t}= {\zeta }_{1}\Delta {y}_{t - 1}+ {\zeta }_{2}\Delta {y}_{t - 2}+ \cdots +{\zeta }_{p - 1}\Delta {y}_{t - p + 1}+ \rho {y}_{t - 1}+ {\varepsilon }_{t}$
${y}_{t}= {\zeta }_{1}\Delta {y}_{t - 1}+ {\zeta }_{2}\Delta {y}_{t - 2}+ \cdots +{\zeta }_{p - 1}\Delta {y}_{t - p + 1}+ \alpha +\rho {y}_{t - 1}+ {\varepsilon }_{t}$
${y}_{t}= {\zeta }_{1}\Delta {y}_{t - 1}+ {\zeta }_{2}\Delta {y}_{t - 2}+ \cdots +{\zeta }_{p - 1}\Delta {y}_{t - p + 1}+ \alpha +{\delta t}+ \rho {y}_{t - 1}+ {\varepsilon }_{t}$
式中: ${\zeta }_{1},{\zeta }_{2},\cdots ,{\zeta }_{p - 1},\rho ,\alpha ,\delta$ 均为常数; ${\delta t}$ 为时间趋势项; $p$ 为滞后阶数; $\left\{{\varepsilon }_{t}\right\}$ 为独立同分布序列,服从均值为 0,方差为 ${\sigma }^{2}$ 的正态分布。
ADF 检验的零假设和备择假设是:
${\mathrm{H}}_{0}: \rho = 1 \\ {\mathrm{H}}_{1}: \rho < 1 $
本文采用式(11)进行 ADF 检验: $\rho < 1$ 时, ${y}_{t}$ 具有非 0 常数均值和有限方差; $\rho = 1$ 时, $\Delta {y}_{t}$ 具有非 0 常数均值和有限方差, ${y}_{t}$ 对应含常数项的随机游走过程,可具有上升或下降的走势。
观察 图 1 右上图,可发现其具有非零均值,且上升趋势不明显,不足以表明其带有时间趋势项。 若采用式(11)对上述序列片段进行检验,则结合上述分析可以预期拒绝零假设。在拐点之后 ( 图 1 右下图),健康指数序列出现了一段快速上升的过程, 若同样采用式 (11) 进行检验, 则可以预期不能拒绝原假设,从而检测出早期故障。
本文所提轴承早期故障预警方法的算法框架如下:
①预处理阶段。采用第 2 节的预处理方法将原始振动信号转换为训练样本;
②迁移学习阶段。按第 2 节提出的目标函数在预训练模型基础上作迁移学习, 得到深度卷积自编码器模型和公共特征中心;
③在线预警阶段。运用已获得的编码器和公共特征中心计算在线样本的健康指标, 形成健康指标序列, 以滑动窗口的形式截取当前序列片段进行 ADF 检验, 最后根据检验结果判断轴承是否出现早期故障。
为验证所提方法的有效性,选择 XJTU-SY 滚动轴承加速寿命试验数据集 [ 12 ] 模拟海上风电机组轴承。该实验通过交流电机转速控制器和液压加载系统改变轴承的转速和所受径向力,设计了轴承在 3 种工况下的加速寿命老化实验。在实验过程中,径向力以水平方向加载在轴承座上,两个单向加速度传感器分别用磁座固定在轴承座的水平和垂直方向上,振动信号采样频率为 ${25.6}\mathrm{{kHz}}$ ,每隔 $1\mathrm{\;{min}}$ 采样 ${1.28}\mathrm{\;s}$ 长度的数据。
采用连续无重叠采样形成样本, 以样本 1024 点数据作 STFT, 长度 64 的汉宁窗作为窗函数, 滑动步长为 32,灰度图样本大小为 ${32}\times {32}$ ; 模型训练批量大小为 64,学习率为 $1 \times {10}^{-4},\lambda = 1 \times {10}^{-3},\alpha = 1$ , $\beta ={0.5}$
为说明所提特征提取模型的性能, 以真阴性 (TN)和真阳性(TP)作为评价指标, 同时用少样本参与训练。采集轴承 1_1 在工况 1 的 400 个正常样本, 训练含 SVDD 的卷积自编码器 (CAE+ SVDD)特征提取模型, 训练 200 周期后计算训练集的健康指标, 将其中的 95 百分位数作为阈值, 训练结束实现 $\mathrm{{TN}}= {35}$$\mathrm{{TP}}= {50}$ 。在工况 2 采集 192 个运行初期样本作迁移学习, 训练含 MMD 正则项的 CAE+SVDD+MMD 模型 200 周期, 提取工况 2 状态特征。为分析所提模型的特征提取性能, 设计如下 3 个参照实验。
①仅用 192 个工况 2 样本, 直接训练 CAE+ SVDD 特征提取模型。
②仅用 320 个工况 2 样本, 直接训练 CAE+ SVDD 特征提取模型。
③ 用 400 个工况 1 样本和 192 个工况 2 样本,训练 CAE+SVDD+MMD 特征提取模型。
实验结果如表表 2所示。
表 2 可知: 卷积自编码器均能准确识别轴承的故障状态, 模型特征提取能力的差异体现在正常状态识别上; 本文所提迁移模型在正确识别正常状态所需训练周期最少, 仅需 160 次即可达到 TN=45,而其他参照实验至少需要 200 次;从正常样本识别的排序看, 迁移学习优于 CAE+ SVDD+MMD,其次是大样本 CAE+SVDD,最差为小样本 CAE+SVDD; 对比参照实验, MMD 正则项有助于模型提取的状态特征更快地向公共特征中心收敛;迁移学习进一步提升了 CAE+SVDD+ MMD 模型的公共特征提取能力, 通过较少样本、 较少算力代价改善了早期故障识别能力。
取 2 个工况的 3 个轴承, 探究所提健康指标在轴承全寿命周期的准确程度。将工况 1 作为源域, 工况 2 作为目标域, 源域和目标域的样本数均为 900 , 轴承的全寿命周期健康指标时间序列如 图 2 所示。
图 2 可知, 大部分轴承的健康指标在早期具有比较平稳的趋势且幅值较低,当发生早期故障后, 健康指标通常会大幅度上升, 出现比较明显的转折点。
为确认健康指标出现突变时是否发生早期故障,以轴承 1_1 和轴承 2_2 发生的外圈故障为例, 用包络谱分析突变点时刻的振动信号。轴承外圈故障特征频率 ${f}_{\mathrm{o}}$
${f}_{\mathrm{o}}= \frac{n}{2}\left({1 -\frac{d}{D}\cos \alpha }\right){f}_{\mathrm{r}}$
式中: $n$ 为轴承滚珠个数; $d$ 为滚珠直径; $D$ 为轴承中径; $\alpha$ 为接触角; ${f}_{\mathrm{r}}$ 为以 $\mathrm{{Hz}}$ 为单位的轴承转速。
可得外圈故障特征频率分别为 ${107.91}\mathrm{\;{Hz}}$${115.61}\mathrm{\;{Hz}}$
图 3 为轴承 1_1 全寿命周期振动信号及幅值增大时刻的包络谱。
图 3 可知, 在第 79 分钟时, 轴承 1-1 振动的幅值开始逐渐增大,查看对应时段的包络谱(虚线表示外圈故障频率及其倍频在频率轴上的位置)可看到, 在这些位置附近出现了显著的频谱分量, 说明出现了外圈故障。
图 4 为轴承 2_2 全寿命周期振动信号及幅值增大时刻的包络谱。
图 4 可知,轴承 2_2 振动幅值增大的时刻在第 49 分钟,同时健康指标也出现了跳变,包络谱中含有接近外圈故障特征频率的谱线及其倍频分量,说明出现了外圈故障。
为了验证所提模型预警性能稳定、误报警率低的优势, 应用所提预警策略对上一节得到的 3 组轴承健康指标进行检测( 图 2 ),其中健康指标已归一化, ADF 滑动窗口长度为 500 , 步长为 1 , 显著性水平 $\alpha$ 设为 $5\%$ 。本文采用 0 和 1 表示当前窗口数据的检测结果, 0 表示无故障, 1 表示有故障,并用针状图表示在图上。从 图 2 灰色区域的针状图可看出, ADF 检验在健康指标突变前没有出现误报警,而在健康指标出现明显上升时,可以准确地给出报警提示。
本文选择了 3 种异常点检测方法用于比较, 方法一为在线检测时将健康指标按时间顺序添加到健康指标序列中,计算其均值 $\mu$ 和标准差 $\sigma$ ,使用 ${3\sigma }$ 准则判断是否报警,其余两种方法为孤立森林和一类支持向量机 [ 6 , 7 ] 图 5 为 3 种方法在两个轴承上的检测结果。
表 3 给出了上述 3 种方法和本文所提方法在两个轴承上的误报警数。
误报警数定义为轴承发生早期故障前, 所用检测方法的报警数。两个轴承的故障时刻分别为第 79 分钟和第 454 分钟。与 图 2 进行对比可知, 3 种方法在轴承 1_1 和 2_1 上具有较多的误报警数,其中方法一的误报警数最多。本文所提方法在所有轴承的健康指标发生显著增大前均没有误报警,并且在故障发生后能持续报警一段时间,说明本文方法具有较高的准确性和稳定性。
为解决海上风电机组工况不一、机组轴承早期故障预警误报警多的问题,本文基于迁移学习的深度卷积自编码器提取状态特征构造了轴承的健康指标, 该健康指标能有效反映轴承在不同工况下状态公共特征的变化。引入时间序列平稳性分析中的 ADF 检验,实现对健康指标序列片段的在线异常检测。本文所提方法无需频繁调整故障阈值, 在公开轴承数据集上与孤立森林、一类支持向量机等方法的对比结果表明,该方法具有误报警少且发生早期故障时能准确稳定报警的特点。
参考文献 引证文献
排序方式:
[1]
British Petroleum. Statistical review of world energy 2020[EB/OL]. https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2020-full-report.pdf2022-07-20. https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2020-full-report.pdf
[2]
Global Wind Energy Council. Global offshore wind report 2020[R]. Brussel: Global Wind Energy Council, 2021.
[3]
Zhang H, Chen X, Du Z, et al. Kurtosis based weighted sparse model with convex optimization technique for bearing fault diagnosis[J]. Mechanical Systems and Signal Processing, 2016, 80: 349-376.
[4]
Sun H, Cao X, Wang C, et al. An interpretable anti-noise network for rolling bearing fault diagnosis based on FSWT[J]. Measurement, 2022, 190: 110698.
[5]
Imaouchen Y, Kedadouche M, Alkama R, et al. A frequency-weighted energy operator and complementary ensemble empirical mode decomposition for bearing fault detection[J]. Mechanical Systems and Signal Processing, 2017, 82: 103-116.
[6]
汤婷, 张岩, 安宗文. 基于孤立森林算法的风电齿轮箱轴承故障检测方法[J]. 轴承, 2022(2): 68-74.
[7]
翟怡萌, 李晶, 程强, 等. 基于OCSVM的火电厂一次风机异常检测[J]. 工业控制计算机, 2019, 32(6): 102-104.
[8]
毛文涛, 田思雨, 窦智, 等. 一种基于深度迁移学习的滚动轴承早期故障在线检测方法[J]. 自动化学报, 2022, 48(1): 302-314.
[9]
赵志宏, 李乐豪, 杨绍普, 等. 一种无监督的轴承健康指标及早期故障检测方法[J]. 中国机械工程, 2022, 33(10): 1234-1243.
[10]
Zhang Z, Deng X. Anomaly detection using improved deep SVDD model with data structure preservation[J]. Pattern Recognition Letters, 2021, 148: 1-6.
[11]
何桂雄, 金璐, 李克成, 等. 基于改进 DaNN 的综合能源系统多能负荷预测[J]. 电力工程技术, 2021, 40(6): 25-33.
[12]
雷亚国, 韩天宇, 王彪, 等. XJTU-SY 滚动轴承加速寿命试验数据集解读[J]. 机械工程学报, 2019, 55(16): 1-6.
2024年第42卷第7期
PDF下载
259
129
引用本文
BibTeX
文章信息
  • 接收时间:2022-07-21
  • 首发时间:2025-07-22
  • 出版时间:2024-07-20
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2022-07-21
基金
作者信息
    1 华南理工大学 广东 广州 510000

通讯作者:

汪隆君(1982-),男,博士,讲师,研究方向为电力系统可靠性与规划。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kzsny/CN/1154429173130056194
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