基于特征筛选和集成学习的轴承故障诊断

Fault Diagnosis of Bearing Based on Feature Selection and Ensemble Learning

为了提高不同工况下的轴承故障诊断准确率,提出了一种基于特征筛选和集成学习的轴承故障诊断方法。考虑到特征向量复杂冗余的问题,结合特征有效性和最大均值差异提出了新的特征评分函数,并在此基础上进一步考虑特征关联度和特征维度,筛选出有利于变工况故障诊断的特征子集。针对单一机器学习模型故障诊断准确率不高的问题,将AdaBoost和Stacking算法相结合构造集成学习故障诊断模型。实验结果表明:筛选出的特征子集在相同分类器下拥有更高的故障诊断准确率;集成学习模型相较于单一模型有更高的故障诊断准确率和鲁棒性。

In order to improve the diagnostic accuracy of bearing failures under different operating conditions,this paper proposes a bearing fault diagnosis method based on feature selection and ensemble learning.Considering the issue of complex and redundant of feature vectors,a new feature scoring function is proposed by combining feature effectiveness and maximum mean discrepancy.Based on this,the feature correlation and feature dimensionality are further considered to select a subset of features that are beneficial for variable operating condition fault diagnosis.Considering the problem of low classification accuracy of a single machine learning model,AdaBoost and Stacking algorithm are combined to construct an ensemble learning fault diagnosis model.Experimental results demonstrate that the selected subset of features exhibits higher accuracy under the same classifier.Moreover,the ensemble learning model demonstrates higher accuracy and robustness compared to a single model.