轴承故障高敏感特征提取与随机森林智能识别  

Robotic Arm Trajectory Multi-Objective Programming Based on Heterogeneous Particle Swarm Algorithm

作　　者：李冬[1] LI Dong(Wuxi Vocational Institute of Arts&Technology,Jiangsu Yixing 214200,China)

机构地区：[1]无锡工艺职业技术学院,江苏宜兴214200

出　　处：《机械设计与制造》2022年第9期157-161,共5页Machinery Design & Manufacture

基　　金：江苏省现代教育技术研究课题(2018-R-64707)。

摘　　要：为了提高轴承故障识别正确率,提出了基于多重分形理论的特征提取方法和改进随机森林的模式识别方法。介绍了多重分形去趋势波动理论,初选了4个多重分形参数作为特征参数;将参数两两组合,使用K均值聚类法进行聚类,依据类内聚集度和类间距离优选了最佳组合作为特征向量。以随机森林算法为基础,提出了舍弃策略和话语权策略进行改进。舍弃策略通过舍弃分类正确率靠后的决策树提高随机森林平均正确率,减小森林的泛化误差;话语权策略通过提高优秀决策树的话语权,放弃了传统算法中的绝对民主,两个改进策略提高了算法模式识别正确率。经实验验证,改进随机森林算法对故障识别正确率为100%,而传统算法识别正确率仅为93.1%,证明了算法改进策略的有效性。To improve accuracy of bearing fault diagnosis,fault feature extraction method based on multifractal theory and mod⁃el diagnosis method based on improved random forest are proposed.Multifractal detrend fluctuation analysis theory is introduced,primarily electing 4 multifractal parameters as feature parameters.One parameter combining with the other,K-means clustering method is used to cluster the parameters combination,and optimal parameters combination is selected by intra-class aggregation and distance among class.The optimal parameters combination is feature vector.Random forest algorithm is improved by abandon strategy and speech right strategy.Abandon strategy means that abandoning latter accuracy decision tree aims to improve average accuracy and reduce generalization error.Speech right strategy means that improve excellent decision tree weight and reject abso⁃lute democracy in traditional random forest.Model diagnosis accuracy of algorithm is improved by the two improving strategies.Clarified by bearing fault experiment,bearing fault diagnosis accuracy by improved random forest is 100%and 93.1%by tradi⁃tional random forest algorithm,which can clarify validity of the improving strategy.

关 键 词：滚动轴承 故障识别 多重分形理论 随机森林算法 话语权 

分 类 号：TH16[机械工程—机械制造及自动化] TH133.33