基于超图相关距离判别投影的轴承故障诊断方法  

Bearing fault diagnosis method based on HCDDP

作　　者：苏树智 张志鹏 SU Shuzhi;ZHANG Zhipeng(School of Computer Science and Engineering,Anhui University of Science&Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《振动与冲击》2023年第23期103-111,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金(52374155);安徽省自然科学基金(2308085MF218);安徽省高等学校科学研究项目(2022AH040113);安徽理工大学研究生创新基金(2022CX2118)。

摘　　要：针对滚动轴承故障数据维度过高以及不同特征属性交错导致的故障分类困难,提出了一种基于超图相关距离判别投影(hypergraph correlation distance discriminant projection, HCDDP)的轴承故障数据降维方法。该方法使用超图结构描述了故障样本间的空间结构关系,并利用轴承故障信号的监督信息构建出类内和类间超图;超图更有效地揭示了故障数据的复杂多重结构,相比传统简单图结构更好的表达了故障样本间的内在性质和多元关系;同时,在超图中提出使用皮尔森相关系数构造了一种新的度量来计算高维流形中样本的测地距离,解决了欧氏距离受故障数据取值范围敏感导致的分类不准确问题;超图相关距离判别投影方法具有的非线性数据高阶关联能力更好的解决了轴承故障的分类代价敏感。该方法在美国凯斯西储大学轴承数据集和西安交通大学轴承数据集上进行了验证。试验结果表明,该方法能够有效利用样本间的多元结构关系和判别信息,提高轴承故障的识别率。Here,aiming at the difficulty of fault classification caused by high dimensionality of rolling bearing fault data and interleaving of different feature attributes,a bearing fault data dimensionality reduction method based on hypergraph correlation distance discriminant projection(HCDDP)was proposed.This method could use a hypergraph structure to describe spatial structural relation among fault samples,and construct intra-class and inter-class hypergraphs using the supervised information of bearing fault signals.Hypergraphs more effectively revealed complicated multiple structure of fault data,and better expressed inherent properties and multivariate relations among fault samples compared to traditional simple graph structure.Meanwhile,building a new metric with Pearson correlation coefficient was proposed in hypergraphs to calculate geodesic distance of samples in high-dimensional manifolds,and solve the problem of incorrect classification caused by sensitivity of Euclidean distance to range of fault data values.The hypergraph correlation distance discriminant projection method with non-linear data high-order correlation ability could better overcome the cost sensitivity of bearing fault classification.The proposed method was verified on bearing data sets of Case Western Reserve University of US and Xi’an Jiaotong University of China.The experimental results showed that the proposed method can effectively utilize multi-variate structural relations and discriminative information among samples,and improve the recognition rate of bearing faults.

关 键 词：故障诊断 滚动轴承 超图学习 维数约简 超图相关距离判别投影(HCDDP) 

分 类 号：TH165.3[机械工程—机械制造及自动化]