用矩阵奇异值分解设计鲁棒故障诊断观测器

基于观测器的故障诊断是控制系统故障诊断的一种重要方法，其关键是对未建模误差和嗓音等未知干扰因素（简称未知输入）的鲁棒性．作者从频域中分析了观测器的误差，得出了对未知输入具有鲁棒性的条件方程式，基于矩阵的奇异值分解理论提出了一种可用于计算机迭代求解的算法，大量的仿真实例验证了本文提出方法的有效性．

A Robust Observer for Fault Diagnosis

Many recent papers deal with fault diagnosis through comparing real syStem with observer system. This paper deals also with this problem but in a different way: more important still, It is much simpler and much more easily implemented.Our idea is essentially the design of a robust observer system. It reduces factors not related to fault, such as noise and unmodeling factors, to below allowable limits. Such reduction is needed for smooth detection of faults, such reduction is achieved through adjusting gain and feedback gain of observer system. Gain and feedback gains must be so adjusted as to satisty condition equation, eq. (12), which is obtained through frequency domain analysis.Some papers also present condition equations, but they are different from eq. (12).Finally we offer simulation results in Figs. 1 through 3. Fig. 1 gives the simulation results of an observer system without reducing factors not related to fault to below allowable limits, Fin. 2 shows that our specially designed observer syStem does nearly completely eliminate the effect of such factors ; Fig. 3 shows that our observer system does give very clear indication of the existence of fault.We emphasize that the simplicity and easy implementation of our observer are due to that we employ the theory of singular value decomposition on solving the condition equation.This design is accomplished automatically with an iterative algorithm.