基于不完全信息的轴承故障聚类识别方法

A Clustering Approach Based on Partial Information for Recognizing Bearing Fault

联合收获机中零部件繁多及滚珠滑失等因素,导致监测信号中轴承组件的特征频率并非总能找到,进而影响了故障诊断的正确率。为了解决该问题,提出了一种基于不完全信息的轴承故障聚类识别方法。该方法将特征频率显著的样本作为先验信息,利用这些信息进行相关成分分析,从而给相关程度高的特征赋予大的权重,然后利用改进的半监督聚类算法对所有样本进行聚类识别。其中,提出了近邻扩展方法对先验信息进行扩充,增加了目标函数惩罚环节对聚类过程予以指导。将所提方法应用于联合收获机的轴承滚珠和外圈故障识别,与其它几种聚类方法相比,故障识别率提高了2.78%~7.22%。

Due to the reasons of too many components in combine harvester and the skid of rolling balls,the characteristic frequencies of bearing assembly in monitoring signals are not always clearly existing,which causes the low accuracy of fault diagnosis. Hence,a clustering approach based on partial information is proposed to tackle this problem. This approach sets these samples with clearly characteristic frequencies as priori information,and then uses them to make relevant component analysis to high weights to relevant dimensions. This approach also design an advanced clustering algorithm to recognize all the samples,wherein an extension strategy based on neighborhood is presented to obtain more priori information,and a penalty step is added to the objective function to guiding the clustering. The fault data on ball and outer race of bearing of a combine harvester is used to validate the proposed approach. The results show that our proposed approach works better than others,where the recognition accuracy is higher than others from 2. 78% to 7. 22%.