基于核schur正交局部Fisher判别的转子故障诊断  

Rotor fault diagnosis based on kernel schur-orthogonal local Fisher discriminant

作　　者：王广斌[1,2] 刘义伦[1] 黄良沛[2] 

机构地区：[1]中南大学机电工程学院,长沙410008 [2]湖南科技大学振动冲击与诊断研究所,湘潭411201

出　　处：《仪器仪表学报》2010年第5期1005-1009,共5页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金资助(50875082)项目

摘　　要：为更好识别转子系统故障,将核映射和向量正交思想引入到局部Fisher判别分析(LFDA)中,提出基于核schur正交局部Fisher判别(KSOLFD)的流形学习算法。首先通过核函数将故障特征信号映射到高维子空间,在该空间重新定义类内散度和类间散度,构建核局部Fisher判别函数,在特征值求解时以schur正交方式找出最优投影向量,进行故障诊断。算法保证了数据降维时重构误差最小,提高了故障诊断效果。实验表明,KSOLFD故障诊断效果相对其他流形学习算法最好。In order to better identify the fault of rotor system, based on kernel mapping and vector orthgonal ideas, a new method based on kernel schur-orthogonal local fisher discrimination （KSOLFD） is proposed. Firstly, the fault feature signals are mapped into high-dimensional space using kernel function, in which within-class and between-class scatters are redefined and local kernel fisher discriminant function is constructed. Then optimal projection vector is found using schur-orthogonal approach and fault diagnosis is realized. This algorithm can minimize data reconstruction error and improve the effect of fault diagnosis. Experiment shows that KSOLFD algorithm has the best effect compared with other manifold learning algorithms in rotor fault diagnosis.

关 键 词：核映射 schur正交 局部Fisher判别 故障诊断 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]