基于岭回归的RVPMCD滚动轴承故障诊断方法

Rolling bearing fault diagnosis method based on RVPMCD

针对多变量预测模型模式识别方法中的最小二乘拟合可能出现病态的问题,提出了基于岭回归的多变量预测模型(Ridge regression-Variable Predictive Model based Class Discriminate,RVPMCD)分类方法,该方法通过引入岭参数,降低其均方拟合误差,减小自变量间复共线性关系对参数估计的影响,改善了原方法中最小二乘回归拟合参数失真的现象,从而有望建立更加准确的预测模型。对滚动轴承的振动信号提取特征值,组成特征向量,采用RVPMCD方法对训练样本建立预测模型,利用RVPMCD所建立的预测模型进行模式识别。实验分析结果表明,基于岭回归的多变量预测模型分类方法可以更有效地对滚动轴承的工作状态和故障类型进行识别。

Aiming at the morbid problem on least-squares fit of variable predictive model, Ridge regression-Variable Predic-tive Model based Class Discriminate（RVPMCD）is put forward. By introducing the ridge parameter on the method, the mean square error on fitting and the effect of multicollinearity on parameter estimation are reduced, and the distortion phenomenon of the least squares regression fit parameter in the original method is improved, therefore, more accurate prediction models can be built up. The feature value of vibration signal of rolling bearings is extracted as feature vector. Then, the RVPMCD method is used to establish prediction model of training samples, and eventually pattern recognition would be carried out by using the established prediction models. The experimental results show that the classification method based on Ridge Regression-Variable Predictive Model can identify work status and fault type of rolling bearings more effectively.