Abstract: To improve the accuracy of early fault identification for rolling bearings,a fault classification model based on hybrid feature extraction was proposed.The model adopted the intra-class compactness and inter-class overlap to identify sub-optimal feature groups,which as the input of Enhanced K-nearest Neighbor (EKNN) classifier.Then the maximum average classification probability was calculated based on distance and density,and the optimal feature set was selected.The fault was detected by classifying the unknown data.The Acoustic Emission (AE) signals of rolling bearings under different working conditions were collected through comparing with other five methods,the proposed method had better performance at early fault diagnosis for rolling bearing.

Key words: acoustic emission signal, enhanced k-nearest neighbor classifier, rolling bearing, early fault classification, fault diagnosis

摘要： 为提高对滚动轴承早期故障识别的精确度,提出一种基于混合特征提取的故障分类模型。该模型利用类内紧致性和类间重叠性识别出次优特征组,作为增强K近邻分类器(EKNN)的输入,并以距离和密度双维度计算,得到最大平均分类精确度,进而输出最优早期故障特征组,对未知数据进行分类来检测故障。实验采集滚动轴承在低速运行下的早期故障声发射(AE)信号,对所提算法以及现有5种算法进行对比分析,验证了其对滚动轴承早期故障诊断具有更好的表现。

关键词: 声发射信号, 增强K近邻分类器, 滚动轴承, 早期故障分类, 故障诊断