针对旋转机械振动故障特征与故障类别间不完全对应问题，以双跨转子系统12个通道的故障信息集合为研究对象，提出一种基于加权核主成分分析的故障敏感特征集合提取方法。通过对每个通道的振动信号进行时域、频域、时频域的特征提取，得到一种描述双跨转子系统的原始故障特征集合。采用多准则特征选择方法对这种原始故障特征集合进行特征属性筛选，得到一种利于故障分类的敏感特征集合。对这12个通道的敏感特征集合进行信息融合处理，可得到一种多通道信息的融合特征向量，利用加权核主成分分析方法提取出融合特征向量中的核主成分。结果表明，这种核主成分能够显示出故障类别间的较显著差异，和具有较好的敏感特征子集寻优能力。该研究为解决好双跨转子系统的故障数据集的类别划分问题，提供了一种新途径。%Aiming at unperfect corresponding relations between fault characteristics and fault categories of rotating machineries,a 12-channel fault information set for a double-span rotor system was taken as a study object,a new method about the feature extraction based on weighted KPCA was proposed.At first,the feature extractions of time domain, frequency domain and time-frequency domain for a single channel vibration signal were done,and the original fault feature set of 12 channels was obtained for the double-span rotor system.Secondly,12 sensitive feature subsets were screened out from the original fault feature set by using the multiple-criterion feature selection method.And then,a fusion feature vector was obtained by fusing the 12 sensitive feature subsets.Finally,the main components of the fusion feature vector were extracted by using the weighted kernel principal component analysis (KPCA).Experimental results showed that this method can find sensitive feature subsets,the kernel main components can reveal the differences among the different fault categories effectively.
展开▼
