摘要
提出了一种非线性流形学习和支持向量机的故障诊断模型。基于机电系统振动信号时域与频域的20个特征参数构建高维特征空间,利用局部切空间排列的非线性流形学习算法提取出隐藏其中的低维流形,网格搜索算法进行维数和邻域点参数的优化,实现高维相空间中局部邻域参数的自适应选取,获得机电系统的故障特征。利用K折交叉验证和一对一法构造支持向量机多类故障分类器,采用径向基核函数支持向量机进行机电系统的故障诊断。应用于转子试验台的3种故障状态的识别并与其他故障诊断方法进行分析比较,结果表明基于局部切空间排列和支持向量机的机电系统故障诊断模型诊断精度可达到96.6667%,可以有效提取故障的敏感特征并解决机电系统故障样本缺乏的问题。
A fault diagnosis model based on nonlinear manifold learning and support vector machine （SVM） is put for- ward. The high dimensional characteristic space is constructed based on 20 characteristic parameters of the vibration sig- nals from machinery and electronic system in time and frequency domains. The lower-dimensional manifold features of the system are extracted using the nonlinear manifold learning algorithm based on local tangent space aligmnent （LTSA）. Grid-search algorithm is used to optimize the number of dimensions and nearest neighborhood point parameters to achieve the self-adoptive selection of local nearest neighborhood parameters in high dimensional phase space. Then the essential fault features of the machinery and electronic system are obtained. The multi-fault classifiers of SVM are constructed using k-fold cross validation and one-against-one method. Radial basis kernel function SYM is adopted to carry out the fault di- agnosis of the machinery and electronic system. The fault diagnosis model was used for the identification of three kinds of fault states of a rotor test platform. The result is compared with those of other fault diagnosis methods. The results demonstrate that the diagnosis accuracy of the machinery and electronic system fault diagnosis model based on local tangent space alignment and support vector machine achieves 96. 6667% ,which is better than those of the other methods. The proposed fault diagnosis model based on LTSA and SVM can effectively extract the fault sensitive features and solve the problem of lacking fault samples in machinery and electronic system.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2012年第12期2789-2795,共7页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(51275052)
北京市自然科学基金重点项目(KZ201211232039)
北京市高等学校人才强教(PHR201106132)资助项目
关键词
机电系统
故障诊断
局部切空间排列算法
支持向量机
网格搜索
machinery and electronic system
fault diagnosis
local tangent space alignment（LTSA）
support vector machine（SVM）
grid search
