A Consideration of Failure Detection in Linear Discrete Systems

This paper is concerned with the failure detection in linear systems whose dynamics change unexpectedly at unknown time. The generalized likelihood ratio test and the sequential probability ratio test have been applied to these problems, under the given threshold value. If we can not find the adequate threshold value, it results in the deterioration of the reliability or the time-lag in the detection of the failure. But the threshold value should be determined from the experimental results because it can not design analytically. Then, we propose the method which detects the failure of the system adaptively, considering the change pattern of the prameters and the statistics of the noise. The parameter estimates are obtained by using the least square method and the failure of the system may be detected on the basis of the performance index considered the residual and the error covariance of estimated prameters. As the change of the system parameters are unknown, it is treated as a step model and a ramp model. It is shown from the results of the digital simulation that the proposed method may be useful for the detection of the system failure.