摘要
针对现有流形学习理论用于旋转机械故障诊断存在识别精度不高的问题,提出基于有监督不相关局部Fisher判别分析(Supervised Uncorrelated Local Fisher Discriminant Analysis,SULFDA)的新型故障诊断方法。首先构造全面表征不同故障特征的时频域特征集,再利用有监督不相关局部Fisher判别分析将高维时频域故障特征集化简为区分度更好的低维特征矢量,并输入到K-近邻分类器中进行故障模式辨识。有监督不相关局部Fisher判别分析在类标签指导下最小化同类流形的离散度并最大化异类流形的离散度来实现类判别,还施加了不相关约束条件使所提取的特征统计不相关,提高了针对旋转机械的故障诊断精度。深沟球轴承故障诊断实验验证了该方法的有效性。
Facing on the crucial problem that the fault diagnosis accuracy of current manifold learning theories for rotating machinery is not high enough,a novel fault diagnosis method based on Supervised Uncorrelated Local Fisher Discriminant Analysis(SULFDA)is proposed in this paper.The time-frequency domain feature set is first constructed to completely characterize the property of each fault.Then,SULFDA is introduced to automatically compress the high-dimensional time-frequency domain fault feature sets of training and test samples into the low-dimensional eigenvectors with better discrimination.Finally,the low-dimensional eigenvectors of training and test samples are input into K-nearest neighbors classifier(KNNC)to carry out fault identification.SULFDA achieves good discrimination ability by minimizing the within-manifold scatter and maximizing the between-manifold scatter under the supervision of class labels.Also,an uncorrelated constraint is put on SULFDA to make the extracted features statistically uncorrelated.Therefore,SULFDA improves the fault diagnosis accuracy for rotating machine.The fault diagnosis experiment on deep groove ball bearings demonstrated the effectivity of proposed fault diagnosis method.
出处
《振动工程学报》
EI
CSCD
北大核心
2015年第4期657-665,共9页
Journal of Vibration Engineering
基金
国家自然科学基金青年基金资助项目(51305283)
国家公派高级研究学者及访问学者(含博士后)项目(201406245021)
高等学校博士学科点专项科研基金资助项目(20120181130012)
关键词
故障诊断
旋转机械
时频域特征集
有监督不相关局部Fisher判别分析
流形学习
fault diagnosis
rotating machinery
time-frequency domain feature set
supervised uncorrelated local Fisher dis-criminant analysis
manifold learning