基于多核非负矩阵分解的机械故障诊断  

The Fault Diagnosis Technology of Mechanical Equipment Based on Multi-Kernel Non-negative Matrix Factorization(MKNMF)

作　　者：杨永生[1,2] 张优云[1] 朱永生[1] 

机构地区：[1]西安交通大学机械工程学院,陕西西安710049 [2]陕西省行政学院计算机系,陕西西安710068

出　　处：《西北工业大学学报》2015年第2期251-258,共8页Journal of Northwestern Polytechnical University

基　　金：国家科技重大专项(2012ZX04005-011)资助

摘　　要：在机械设备故障诊断研究领域中,系统采集的原始监测数据经过处理得到的结果往往是数据量很大,维数很高的图像数据,因此,从高维图像中获取敏感特征是当前故障诊断领域中面临的一项关键技术。本文提出了基于多核非负矩阵分解的机械设备故障诊断方法,该方法克服了传统故障诊断需对机械设备信号进行特征提取而造成信息丢失,通过应用多核非负矩阵分解方法进行降维,然后结合多核支持向量机实现对降维后的数据直接进行识别。实验证明该方法可降低原始数据特征的维数,提高分类运算的效率以及故障诊断的识别率。In the fault diagnosis field of mechanical equipment,the result of analyzing the collected monitoring data from the equipment is often the high dimensionality of images which contain mass data; so the method of extracting sensitive feature from the high-dimensional information or image is a key technology. We present a new method for fault diagnosis of mechanical equipment based on Multi-Kernel Non-negative Matrix Factorization（MKNMF）,which overcomes the defect that the traditional fault diagnosis of mechanical equipment requires signal feature extraction this defect causes loss of information; we reduce dimensions for high dimension information through applying Multi-Kernel Non-negative Matrix Factorization method and then distinguish the dimensionality reduction data with Multi- Kernel Support Vector Machine（MKSVM）. The experiments and their analysis show preliminarily that this method can reduce the dimensions of the original monitored data and improve the recognition rate of machine fault diagnosis.

关 键 词：多核非负矩阵分解 支持向量机 故障诊断 数据降维 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]