基于SILPDA的旋转机械故障诊断方法

Fault diagnosis method of rotating machinery based on SILPDA

针对故障特征集因“维数灾难”导致的故障分类困难现状,提出了一种基于强化内蕴局部保持判别分析(strengthened intrinsic local preserving discriminant analysis, SILPDA)的故障特征集降维算法。该算法将强化的多流形内蕴模型与局部相似度矩阵融入目标函数的构建中,期间充分考虑了数据集的多流形结构特征并且保留了样本的局部结构信息,使降维后的低维特征子集易于实施分类运算,继而实现提高故障辨识精度的效果。利用转子试验台振动信号集合构建的原始故障特征集对算法性能进行了验证。结果表明,该算法能够从原始故障数据集中提取出利于实施分类运算的敏感特征子集,这些特征子集将会使不同故障类别之间的边界变得更加清晰,最终相较于局部保持投影(locality preserving projections, LPP)、线性判别分析(linear discriminant analysis, LDA)、局部边缘判别投影(locality margin discriminant projection, LMDP)等算法可实现更好的故障辨识效果。对于提高旋转机械大数据资源的价值密度,该算法提供了一种优化数据结构模型的理论依据。

Aiming at the difficulty of fault classification caused by “dimension disaster” in fault feature sets, a dimension reduction algorithm for fault feature sets based on strengthened intrinsic local preserving discriminant analysis(SILPDA) was proposed. The algorithm integrated the enhanced multi manifold intrinsic model and local similarity matrix into the construction of the objective function. During this period, the multi manifold structure characteristics of the data set were fully considered and the local structure information of the sample was retained, so that the low dimensional feature subset after dimensionality reduction is easy to implement classification operation, and then achieve the effect of improving the accuracy of fault identification. The performance of the algorithm was verified by using the original fault feature set constructed from the vibration signal set of a rotor test-bed. The results show that the algorithm can extract sensitive feature subsets that are conducive to the implementation of classification operation from the original fault data set. These feature subsets will make the boundary between different fault categories clearer. Finally, compared with locality preserving projections(LPP), linear discriminant analysis(LDA), locality margin discriminant projection(LMDP) and other algorithms, the algorithm can achieve better fault identification effect. For improving the value density of rotating machinery big data resources, this algorithm provides a theoretical basis for optimizing the data structure model.