基于近邻元分析的滚动轴承故障诊断方法

Fault diagnosis method for rolling bearing based on neighborhood component analysis

提出基于近邻元分析(Neighborhood Component Analysis,NCA)的滚动轴承故障诊断模型。利用NCA技术对滚动轴承信号的时、频域特征向量降维,并对降维后向量分类,成功区分滚动轴承四种状态。通过Fisher判别函数定量分析目标维数对NCA降维效果影响,确定最佳特征约简目标维数。为突出NCA方法优势,将NCA与PCA(Principle Component Analysis)两种不同降维方法进行对比。实验结果表明,NCA作为监督式降维方法,其聚类效果好于PCA。

A novel fault diagnosis method for rolling bearing was proposed based on neighborhood component analysis（NCA）.NCA was used to reduce the dimensions of input features in both time and frequency domains.Then the classification was performed.Fisher evaluation function was applied to select the proper object dimension.In order to show the advantages of the proposed method,the classification results based on PCA and NCA were compared.The experiment shows that as a supervised dimension reduction method,NCA performs better than PCA.