为了克服传统主元分析(Principal component analysis,PCA)因模式复合现象而无法进行多故障诊断和诊断结果难以解释的不足,本文引入指定元分析(Designated component analysis,DCA)的思想,建立DCA多故障诊断理论的空间投影框架,从而把异常检测问题转化为将观测数据向故障子空间投影后投影能量的显著性检测问题.在确定系统存在异常的情况下,再将观测数据向故障子空间中各故障模式方向分别进行投影,根据投影能量的显著性进行多故障诊断.并利用正交补空间构造法证明了基于非正交模式指定元分解形式的可行性和收敛性,建立了一种逐步DCA多故障诊断方法以解决指定模式非正交情况下的多故障诊断问题.包含5种共存故障的观测数据的仿真研究验证了新方法的有效性.
As it can avoid pattern compounding problem of principal component analysis (PCA), designated component analysis (DCA) is introduced to implement multiple faults diagnosis of a multivariate system. A projection frame, which is the theory foundation for DCA based multiple faults diagnosis, is established in this paper. Under the DCA projection frame, anomaly detection and fault diagnosis problem is transformed into the significance detection problem for projection energy of the observation data projected to fault subspace spanned by fault patterns defined in advance. Then, for the case when designated patterns are not orthogonal to each other, a progressive DCA diagnosis method is developed by grouping the designated patterns into several orthogonal subgroups to the observation data or the residual of the previous DCA process. Simulation for data involved 5 faults shows the efficiency of the progressive DCA method for multiple faults diagnosis.