有监督LLTSA特征约简旋转机械故障诊断

Rotating machinery fault diagnosis with supervised-linear local tangent space alignment for dimension reduction

线性局部切空间排列(LLTSA)为无监督特征约简方法,对多域故障特征集进行维数约简,会导致故障解耦不完全、故障间仍然存在混叠。针对这个问题,提出有监督线性局部切空间排列(S-LLTSA)特征约简方法,将类判别信息融入特征约简过程,实现了数据集本征结构与类判别信息的有机结合,可提取出最优低维敏感故障特征向量;并通过自适应近邻分类器(ANNC)来构建故障特征向量与故障类别的对应关系。S-LLTSA特征约简有效地增加了故障特征的可辨识性,而ANNC具有优异的模式辨识能力,进一步提高了故障诊断的精度。齿轮箱故障模拟实验验证了提出的旋转机械故障诊断方法的有效性。

Linear local tangent space alignment （LLTSA）is an unsupervised feature reduction method;this method leads to the incom-plete fault decoupling and remaining overlaps between faults when it is used to perform LLTSA on multi-domain feature set for dimension reduction.Aiming at the shortcoming of LLTSA,supervised-linear local tangent space alignment （S-LLTSA）for feature dimension re-duction is proposed.In the method of S-LLTSA,the combination of data set intrinsic structure and class discrimination information is re-alized through integrating the class discrimination information into the feature reduction process.As a result,the optimal sensitive low-di-mensional fault feature vectors are obtained.And then,the corresponding relationship between fault feature vectors and fault classes are established using adaptive nearest neighbor classifier.Dimension reduction with S-LLTSA can effectively increase the discrimination of fault features;and what is more,adaptire nearest neighbor classifier （ANNC）can further improve the accuracy of fault diagnosis with its excellent pattern recognition ability.At last,the effectiveness of the proposed method was verified through the gearbox fault simulation experiment.