基于流形学习与一类支持向量机的滚动轴承早期故障识别方法

Incipient Fault Recognition of Rolling Bearings Based on Manifold Learning and One-class SVM

提出了一种基于流形学习与一类支持向量机的轴承早期故障识别方法。首先提取轴承信号的时域参数构成原始特征样本空间;然后采用基于拉普拉斯特征映射算法(Laplacian eigenmap,LE)的流形学习方法对特征样本进行特征压缩,提取出敏感的故障特征;最后采用一类支持向量机对各状态实现分类识别。利用实测的滚动轴承故障数据对算法进行了验证,并将LE方法与主成分分析(PCA)方法进行了比较,结果证明该方法可行。

A method was presented for incipient fault recognition of rolling bearings, which was based on manifold learning and one-class SVM. Firstly,the original feature space was constructed with the domain parameters of bearing signals, and then the LE was used to compress the feature samples and acquire the sensitive fault features, Finally the classification and recognition of all status were implemented with one-class SVM. Besides, with the actual fault data of rolling bearings the method was confirmed and the feasibility was indicated from the comparison of LE and principal component analysis.