一种基于聚类的滚动轴承故障诊断方法

近年来,机器学习技术在故障智能诊断领域得到了广泛的应用,聚类作为最主要的无监督学习技术在基于机器学习的故障智能诊断中占有重要的地位。滚动轴承故障诊断中,传统的频谱分析法通常采用共振解调技术,但当内圈、滚动体或多点故障时,产生复合调制,从解调谱线很难分辨故障类型。针对此,提出了一种新的基于模糊聚类的滚动轴承故障诊断方法,该方法以模糊Fisher准则为聚类目标,通过对待测样本与已知状态样本数据聚类,求得待测样本隶属度,进而判断滚动轴承的故障类型。实验结果表明该方法是有效的。

A rolling bearing fault diagnosis method based on clustering

In recent years,machine learning techniques have been widely used in fault intelligent diagnosis field.Clustering,as a major unsupervised learning technology,occupies an import position in failure intelligent diagnosis based on machine learning.In rolling bearing fault diagnosis,the traditional spectrum analysis method usually adopt the resonant demodulation technology,but when the inner circle,rolling body or multi-point faults produce composite modulation,it is difficulty to identify the fault type from demodulation spectral lines.According to this,a novel rolling bearing fault diagnosis method based on clustering is proposed.It uses fuzzy Fisher criterion as its clustering goal.Through clustering on test samples and the known samples,the memberships of test samples are obtained.From these,the rolling bearing fault type can be judged.Experimental results show that this method is effective.