一种数据驱动的多故障诊断方法研究

On Data Driven Multiple Fault Diagnosis Method

结合统计建模与物理建模的指定元分析(DCA)有效地避免了主元分析的模式复合问题,从而能用来进行多故障诊断。针对非正交指定模式的处理问题,基于模式分组思想给出一种逐步DCA多故障诊断法,把常见变化模式分为几个正交模式组,然后关于各组指定模式,分别对所得观测数据阵或上一步DCA所得残差矩阵做DCA,逐步诊断各组中的故障是否发生。针对包含6种共存故障的观测数据的仿真研究表明,逐步DCA多故障诊断算法可有效地进行多故障诊断,且无需人为解释诊断结果的物理意义。

Designated component analysis（DCA） which can combine statistic pattern building and physics pattern building can effectively avoid pattern composing problem so as to make fault diagnosis.For orthogonal variation pattern must be designed in advance,a progressive DCA multi-fault diagnosis method is given based on the idea of pattern grouping.Common fault patterns are grouped into orthogonal subgroups,then according to each designated pattern,DCA is implemented on the observation data or the residual.And significance of each fault pattern is computed to identify whether the corresponding fault will occur.Simulation result involving observation datum of 6 coexistent faults show that the DCA multifault diagnosis algorithm can effectively make multi-fault diagnosis without explanation of physical meaning of the diagnosing result.