多准则融合敏感特征选择和自适应邻域的流形学习故障诊断

Manifold learning method for fault diagnosis based on sensitive feature selection with multi-criteria evaluation sequences and adaptive neighborhood

针对流形学习故障诊断中非敏感特征干扰和邻域大小难以确定的问题,提出了基于DSm T多准则融合特征选择和局部集聚系数自适应邻域的流形学习故障诊断方法。利用多种特征评价准则对原始高维特征进行排序,通过DSm T证据理论对各评价序列进行融合,再根据融合序列选择敏感特征构成优化高维特征集;采用基于局部集聚系数的自适应正交邻域保持嵌入流形学习进行维数约简,得到低维特征集;最后输入到K最近邻分类器进行故障识别。轴承振动故障实验表明了本文所提方法的有效性。

In order to solve interference of non-sensitive features and the neighborhood size of the manifold learning, in the present paper, a novel manifold learning method for mechanical fault diagnosis based on feature selection with Dezert-marandache theory（DSm T） and adaptive neighborhood with local cluster coefficient is proposed. Multi feature evaluation criterias are used to sort the original high-dimensional features, a fusion sequence by DSm T is used to extract optimal subset. The adaptive neighborhood of orthogonal neighborhood preserving embedding（ONPE） with local cluster coefficient is used to reduce the high-dimensional set to the low-dimensional compressed sensitive feature subset. Then, fault is identified with feeding the feature subset into the k nearest neighbor classification（KNNC）. At last, the validity of the proposed method is verified with fault diagnosis tests of bearings.