基于多尺度极差熵和专家森林的轴承故障诊断  

Fault diagnosis of bearing based on multi-scale rang entropy and expert forest

作　　者：张书锋[1] 陈雪勤[2] ZHANG Shu-feng;CHEN Xue-qin(School of Information Engineering,Suzhou Industrial Park Institute of Services Outsourcing,Suzhou 215123,China;School of Electronic Information,Soochow University,Suzhou 215006,China)

机构地区：[1]苏州工业园区服务外包职业学院信息工程学院,江苏苏州215123 [2]苏州大学电子信息学院,江苏苏州215006

出　　处：《机电工程》2022年第1期47-52,共6页Journal of Mechanical & Electrical Engineering

基　　金：苏州市教育科学“十三五”规划课题资助项目(192012409)。

摘　　要：针对轴承故障的敏感特征提取与模式诊断问题,提出了一种基于多尺度极差熵的故障特征提取方法和基于专家森林算法的轴承故障识别方法。首先,介绍了轴承的3种故障模式和相应的故障特征频率,引入了多尺度理论,计算了轴承振动信号的多尺度极差熵,并将其作为特征向量;然后,研究了特征向量的主成分提取法,有效降低了特征向量维度,并提取出了高敏感特征;最后,在随机森林算法中,依据决策树预测试准确率为决策树赋予了专家权值属性,提出了专家森林算法,并将专家森林算法应用于轴承故障诊断实验中。研究结果表明:使用降维特征加随机森林的故障诊断准确率比降维前特征加随机森林的诊断准确率高了17.07%,降维特征加专家森林算法的故障诊断准确率比降维特征加随机森林高了3.47%;该结果也验证了多尺度极差熵特征提取方法与专家森林故障模式识别的有效性。Aiming at the problem of sensitive feature extraction and pattern diagnosis of bearing fault,a fault feature extraction method based on multi-scale range entropy and a fault recognition method based on expert forest algorithm were proposed.Firstly,three fault modes and corresponding fault characteristic frequencies of bearings were introduced.The multi-scale theory was introduced,and the multi-scale range entropy of bearing vibration signal was calculated as the feature vector.Then the principal component extraction method of feature vector was studied,which effectively reduces the dimension of feature vector and extracts highly sensitive features.Finally,in the random forest algorithm,the expert weight attribute was given to the decision tree according to the pre-test accuracy of the decision tree,the expert forest algorithm was proposed,and the expert forest algorithm was applied to bearing fault diagnosis.The results show that the fault diagnosis accuracy of dimension reduction feature plus random forest is 17.07%higher than that of feature plus random forest before dimension reduction,and the fault diagnosis accuracy of dimension reduction feature plus expert forest algorithm is 3.47%higher than that of dimension reduction feature plus random forest.The effectiveness of the multi-scale range entropy feature extraction method and expert forest fault pattern recognition is verified.

关 键 词：滚动轴承 故障诊断 多尺度极差熵 专家森林算法 主成分分析 高敏感特征 

分 类 号：TH133.33[机械工程—机械制造及自动化]