A two stage real time fault monitoring system

In the present paper, a two-stage fault monitoring scheme is proposed and evaluated for discrete linear stochastic systems in the event of sensor noise degradation. In the first stage, fault detection and partial isolation is achieved using simple statistical tests based on the failure effect on the joint pdf of the Kalman-Bucy filter innovations. Complete fault identification (fault size, fault location, time of occurrence) is performed by the subsequent use of generalized likelihood ratio test (GLR). The analysis is verified by computer simulation of a first order discrete stochastic system.