Abstract:

The fault diagnosis of rolling bearings was a pattern recognition problem essentially, variable predictive model-based class discriminate(VPMCD) was a new pattern recognition method, in essence, through the inner relations among these features, a mathematic predication model that could be established and used to predict the features of those diagnosed bearings and thus to achieve the goal of pattern recognition. But the VPMCD classification method could only use the regression model to predict, so it would reduce the prediction accuracy when the relation of features was complicated. Aiming at this defect, a KVPMCD pattern recognition method was presented. Kriging model was composed of regression model and correlation model, and the latter was local deviation which was created on the basis of global model, and it could reveal the spatial correlation with features exactly, so as to make up for the shortcomings of simple regression model in the original VPMCD.The analysis results of the UCI standard data and the experimental data of rolling bearing show that compared to the original VPMCD, KVPMCD pattern recognition method can identify the working states and fault types of rolling bearings effectively.

Key words: Kriging model, Kriging variable predictive mode based class discriminate(KVPMCD), rolling bearing, fault diagnosis

摘要：

滚动轴承的故障诊断本质上是模式识别的问题,多变量预测模型(VPMCD)是一种新的模式识别方法,其实质就是通过特征值之间的相互内在关系建立数学模型,并根据数学模型对被诊断轴承的特征值进行预测从而达到模式识别的目的。但是VPMCD分类方法中单纯采用回归模型进行预测,因此当故障特征值之间关系较为复杂时将导致预测精度降低。针对这一缺陷,提出了基于Kriging函数的多变量预测模型(KVPMCD)模式识别方法。Kriging模型由回归模型和相关模型组合而成,其中,相关模型是在全局模型的基础上创建的局部偏差,它恰恰可以揭示特征值之间的空间相关性，从而弥补原VPMCD中单纯采用回归模型的缺点。对UCI标准数据以及滚动轴承实测数据的分析结果表明，KVPMCD模式识别方法比原VPMCD方法可以更加有效地识别滚动轴承的工作状态和故障类型。

关键词: Kriging模型, KVPMCD, 滚动轴承, 故障诊断