基于多维度信息融合评判方式的滚动轴承故障特征选择方法  

Fault feature selection method for rolling bearings with multidimensional information fusion judging method

作　　者：吴斌鑫 刘美 周正南 吴猛[1] 张斐 WU Bin-xin;LIU Mei;ZHOU Zheng-nan;WU Meng;ZHANG Fei(College of Information and Control Engineering,Jilin Institute of Chemical Technology,Jilin 132022,China;School of Automation,Guangdong University of Petrochemical Technology,Maoming 525000,China;School of Mechanical Engineering,Dongguan University of Technology,Dongguan 523419,China;Hunan Provincial Key Laboratory of Health Maintenance for Mechanical Equipment,Xiangtan 411100,China)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132022 [2]广东石油化工学院自动化学院,广东茂名525000 [3]东莞理工学院机械工程学院,广东东莞523419 [4]机械设备健康维护湖南省重点实验室,湖南湘潭411201

出　　处：《机电工程》2023年第6期825-834,共10页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金面上基金资助项目(62073091);广东省高校重点领域专项(2020ZDZX3042);广东省普通高校特色创新项目(2017KTSCX176);广东省普通高校机器人与智能装备重点实验室项目(2017KSYS009);东莞理工学院机器人与智能装备创新中心项目(KCYCXPT2017006);机械设备健康维护湖南省重点实验室开放基金资助项目(21903)。

摘　　要：针对多传感数据流及故障特征构建所带来的高维数据的时间复杂度高与数据冗杂问题,提出了一种基于多维度信息融合评判方式的轴承故障特征选择方法。首先,以随机森林、Spearman相关性分析作为基点,并结合门控循环单元(GRU)、差分整合移动平均自回归模型(ARIMA)对各特征做出了初步评价;其次,引入了新评价函数,融合了各部分初步评价信息,剔除了尾部特征并逐次迭代,选择了低冗余且具有较好分类效果的特征子集;最后,以美国凯斯西储大学轴承数据为例,对基于多维度信息融合评判方法与基于随机森林估计器的递归特征消除(RFE-RF)、最大相关最小冗余(mRMR)相比较,以分类准确率作为评估指标,验证了模型的效果。研究结果表明:该方法能在保持97.5%准确率的情况下,得到较少的特征子集,提升了计算效率;该模型能够为滚动轴承故障特征的选取提供借鉴。A bearing fault feature selection method with a multidimensional information fusion judging approach was proposed to address the high time complexity of high-dimensional data and data redundancy with multi-sensing data streams and fault feature construction.Firstly,the random forest and Spearman correlation analysis were used as the base point,and combining with gate recurrent unit(GRU)and auto regressive integrated moving average(ARIMA)model,the preliminary evaluation of each feature was made.Then,a new evaluation function was introduced to fuse the preliminary evaluation information of each part,eliminate the tail features and iterate one by one.A subset of features with low redundancy and better classification effect was selected.Finally,the proposed method was compared with recursive feature elimination based on random forest(RFE-RF)and maximum-relevance minimum-redundancy(mRMR)using Case Western Reserve University bearing data as an example,and the classification accuracy was used as an evaluation index to validate the model effect.The research results show that the method can obtain fewer feature subsets and improve computational efficiency while maintaining an accuracy of 97.5%.The model can provide reference for the accurate selection of rolling bearing fault characteristics.

关 键 词：数据冗余 随机森林 相关性分析 门控循环单元 信息融合 轴承数据 

分 类 号：TH133.33[机械工程—机械制造及自动化]