一种结合谱聚类与关联规则的轴承故障诊断方法

A Bearing Fault Diagnosis Method Combining Spectral Clustering and Association Rules

摘要:

针对现阶段机械设备轴承故障诊断方法难以挖掘隐含特征、诊断精准度低等问题，将谱聚类（spectral clustering，SC）算法与关联规则算法Apriori相结合，提出SC-Apriori算法。首先根据美国西储大学轴承数据中心网站公开发布的轴承故障数据集，选取0负载下的数据，计算得到滚动轴承振动信号的9个时域特征和3个频域特征；其次使用Pearson相关系数进行特征筛选，留下9个有效特征，再利用SC-Apriori算法挖掘出训练数据集中轴承不同特征数据之间的关联关系，并引入提升度来去除冗余的关联规则，进而构建一个规则库；再将测试数据进行处理，并与已建立的规则库进行比对，根据匹配率来判断其故障类型。在测试数据上的实验结果表明，与已有算法相比，本文设计的SC-Apriori算法挖掘出的规则数量大幅减少，匹配速度更快，且匹配效果更好。

Abstract:

The SC-Apriori algorithm was proposed by combining the spectral clustering (SC) algorithm with the association rule algorithm Apriori to address the problems of difficulty in mining the implicit features and low diagnostic accuracy of the existing mechanical equipment bearing fault diagnosis methods. Firstly, based on the bearing fault dataset publicly released on the website of the Bearing Data Centre of Western Reserve University, the data under 0 load were selected and nine time-domain features and three frequency-domain features of the rolling bearing vibration signal were calculated; secondly, the Pearson correlation coefficient was used to filter the features, leaving nine effective features, and then the SC-Apriori algorithm was used to mine different features of the bearings in the training dataset. The association relationship between the data and the introduction of boosting to remove the redundant association rules, and then construct a rule base; then the test data are processed and compared with the established rule base to determine their fault types according to the matching rate. The experimental results on the test data show that the SC-Apriori algorithm designed in this paper mines a significantly reduced number of rules, matches faster and has better matching effect compared with existing algorithms.