关键词: 递归主元分析, 自适应故障监测, 可变遗忘因子, 部分奇异值分解, 时变过程

Abstract:

In order to avoid false alarms for time-varying process and missed alarms for weak fault, an improved recursive principal component analysis (RPCA) method based on variable forgetting factor was proposed for adaptive fault monitoring. A new variable forgetting factor style was introduced for online update of the principal component model, and different forgetting factors were set for different parameters. The loading matrix and eigenvalue matrix were updated by applying partial singular value decomposition (PSVD) method to the recursive decomposition of correlation matrix. In addition, a recursive updating method of control limit was proposed to update control limit adaptively. Monitoring results of the working process of a radar transmitter demonstrate that the improved RPCA method could capture the variation of process adaptively to detect fault, and could reduce both false alarms for normal working condition adjustment and missed alarms for weak fault.

Key words: recursive principal component analysis, adaptive fault monitoring, variable forgetting factor, partial singular value decomposition, time-varying process