基于半监督聚元自组织映射的齿轮早期故障检测.pdfVIP

66 2011 : 1004- 2539( 2011) 11- 0066- 05 ( , 510640) 自组织特征映射应用于齿轮早期故障检测时, 常常导致训练时间长检测精度低故障分类 果不直观等问题提出了基于LDA( 线性判别分析) 与半监督聚元自组织映射的故障检测方法, 首先 利用LDA 对故障特征集进行降维处理, 然后再利用半监督聚元自组织网络对降维后的特征子集进行分 类并将 果可视化, Iris 数据集的仿真 果验证了该方法的有效性通过齿轮早期故障模拟试验对提出 的诊断方法进行了验证, 果表明, 该方法能有效地实现齿轮早期故障检测 半监督学习 聚元自组织特征映射 早期故障诊断 特征选择 Gear Incipient Fault Diagnosis based on Semi- supervised Grouping Neuron Self- organized Map Fan Fan Xu Yabing ( Automotiv Engin ring Institut , Guangzhou Automobil Group Co. , Ltd. , Guangzhou 510640,China) Abstract S lf- organizing f atur map is oft n limit d by long training tim , low diagnosis accuracy and bad classification visualization wh n it is appli d in g ar fault diagnosis. A nov l m thod of s mi- sup rvis d LDA- GN- SOM (Lin ar Discriminativ Analysis Grouping N uron S lf- organiz d Map) is propos d for fault diagnosis. Firstly, th fault of original f atur spac is r duc d by using LDA, and th n th r duc d f atur subs ts ar input into th s - mi- sup rvis d GNSOM n twork for classification and visualization. Th ff ctiv n ss of this m thod is v rifi d by us- ing Iris datas t. And th g ar incipi nt fault xp rim nts ar also conduct d to d monstrat that th n w approach is f- f ctiv for g ar incipi nt pitting d t ction. Key words S mi- sup rvis d l arning Grouping N uron S lf- organizing Map Incipi nt fault diagnosis F atur s l ction SOM Kohon n 1982 0 [4] , , , SOM