基于索引冗余字典的轴承故障组稀疏分类方法研究  

Group sparse representation-based classification method of bearing faults based on index redundant dictionary

作　　者：邓韬 林建辉[2] 黄晨光[2] 靳行[2] DENG Tao;LIN Jianhui;HUANG Chenguang;JIN Hang(College of Electrical & Information Engineering,Southwest Minzu University,Chengdu 610041,China;State Key Laboratory of Traction Power,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南民族大学电气信息工程学院,成都610041 [2]西南交通大学牵引动力国家重点实验室,成都610031

出　　处：《振动与冲击》2019年第7期1-8,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金重点项目(61134002)

摘　　要：基于声发射信号的高速列车轮对轴承早期故障状态诊断和分类复杂性高,常用的人工神经网络及支持向量机方法在参数设置与多分类问题上存在困难。组稀疏分类(GSRC)仅通过超完备字典下稀疏重构即可实现理想的多分类,在图像、语音分类中成为热点。为将GSRC用于轴承故障识别,设计了一种带索引的复合故障冗余字典,利用样本信号多尺度排列熵构成索引字典的小体积优势预先匹配来缩小故障类范围,以邻近梯度法和最优一阶加速的组LASSO约束优化算法来提高收敛性和计算速度;采用改进EEMD结合变分模态分解自适应的获得各故障类初始原子,以保留故障的非线性特征,同时提出一种原子区间平移稀疏编码方法(Interval Translation Sparse Coding, ITSC)放宽了样本数据截取要求,原子有更好的紧凑性与稀疏性;对七类轴承缺陷试验台跑合声发射信号进行分类,验证了该方法的性能。To diagnose and classify early fault states of high-speed train wheel pairs’ bearings based on acoustic emission signals are very complicated, and to set parameters and cope with multi-classification problems are difficult with commonly used artificial neural network and SVM. The group sparse representation-based classification(GSRC) method can be used to realize ideal multi-classification through sparse reconstruction under a super-complete dictionary, and it becomes a hot spot in image and speech classification. Here, a composite bearing faults redundant dictionary with indexes was designed for the GSRC method to be used in bearing fault diagnosis. Small volume advantage pre-allocation of index dictionary constructed with multi-scale permutation entropy of sample signals was used to narrow the range of fault classification. The neighborhood gradient method and the optimal first order accelerated least absolute shrinkage and selection operator(LASSO) constrained optimization algorithm were used to improve convergence and computation speed. The improved EEMD method combined with the variational mode decomposition(VMD) was used to adaptively obtain initial atoms of various fault classes, and keep faults’ nonlinear features. An interval translation sparse coding(ITSC) method was proposed to relax requirements of sample data interception to make atoms have better compactness and sparseness. Classification of running acoustic emission signals was conducted for 7 kinds of bearing defect test bench to verify the effectiveness of the proposed method.

关 键 词：轴承故障 组稀疏分类 声发射 VMD 

分 类 号：TN911.7[电子电信—通信与信息系统] TH17[电子电信—信息与通信工程]