
ISSN10003762 CN411148/TH
轴承2016年12期 Bearing 2016,No. 12
基于NSP和SVM的滚动轴承故障诊断方法
吴春光,王建朝,化麒
(中国人民解放军第五七一五工厂,河南洛阳
471000)
39 42,55
摘要:针对滚动轴承故障诊断中难以获得大量典型的故障样本,以及统计多数对于滚动轴承故障的多分类效果不理想问题,提出了一种基于零空间追踪算法和支持向量机的故障模式识别方法。首先,根据轴承振动模型估计出相应的微分算子;然后,利用上述微分算子将轴承故障信号分解为一系列具备轴承故障特征的窄带信号之和;最后,计算各窄带信号的统计参数,构造特征向量并利用SVM进行故障模式识别。与传统SVM的对比分析试验结果表明:该方法的诊断准确率高达95%,比传统SVM提高了15%,可有效实现滚动轴承故障的模式识别
关键词:滚动轴承;故障诊断:零空间追踪算法;支持向量机;故障识别
中图分类号:TH133.33;TN911.7
文献标志码:B
文章编号:10003762(2016)12003904
FaultDiagnosis ofRollingBearingsBasedonNull-SpacePursuit
Algorithm andSupportVectorMachine
Wu Chunguang,Wang Jianchao,Hua Qi
( The No. 5715 Factory of the Chinese Peoples Liberation Army , Luoyang 371000, China
Abstract : Aiming at the difficulty to obtain enough samples and statistical parameters in roller bearing fault diagnosis, especially the statistical parameters being not ideal for multi classification of rolling bearing faults, a new method of fault diagnosis and pattem recognition for rolling bearing is proposed based on the algorithm of null space pursuit and SVM ( support vector machine). First,the null space differential operator based on the vibration signal model for the faulty bearing fault is established. Then , by using the null space differential operator proposed, vibration signals to be analyzed are decomposed to a series of narrowband signals. Finally,the statistical paramelers of the narrow band signal are calculated to construct the feature vector and the SVM is used to identify the fault patterns. Compared with tradi-tional SVM,the results show that the diagnostic accuracy of this method is as high as 95% , which is 15% higher than
that of traditional SVM, which can effectively realize the patterm recognition of rolling bearing fault. Key words : rolling bearing;fault diagnosis; NSP algorithm ; SVM; fault patem recognition
滚动轴承的故障诊断过程主要分为故障的特征信息提取和状态识别2个部分,即在提取故障的特征信息后,利用这些信息对轴承故障类型或部位进行识别。在识别过程中,故障信息利用的越充分,故障识别就会越准确、可靠,也就是说故障的特征提取是后续故障识别的基础,故障识别的效果依赖于故障特征的提取[1]。
实际工况中,传感器所采集的轴承振动信号
往往含有较强的背景噪声,故障特征没于噪声中,给轴承的故障特征信息提取造成困难。零空
收稿日期:201604-22;修回日期:2016-0704 万方数据
间追踪(NullSpacePursuit,NSP)算法是在EMD算法基础上提出的一种基于局部窄带信号和算子理论的自适应分解算法[2],其核心思想是局部窄带信号在奇异局部线性算子作用下“消失”,因此可以将奇异局部线性算子作用到信号上以抽取信号的局部窄带分量,并将得到的局部窄带信号作为基本信号进行叠加从而逼近原始信号,以实现信号的自适应分解,具有良好的鲁样性、自适应性等优点,在图像处理、轴承故障诊断领域都有相关应用[2-8]
目前,在故障诊断的状态识别中,支持向量机(SupportVectorMachine,SVM)是一种基于结构风