Fault Diagnosis in Dynamic Systems Via Kalman Filter Innovation Sequence

Abstract

Abstract In this paper, a real-time approach to detect and isolate the faults affecting the mean, and the covariance matrix of the Kalman filter innovation sequence, is presented. As monitoring statistics, the ratio of two quadratic forms of which matrices are theoretic and selected covariance matrices is used. A theorem is given to show that the arguments of the optimal quadratic form maximize the above statistics, and thus they are determined to detect and isolate faults in the sensors rapidly. The longitudinal dynamics of an aircraft control system, as an example, is considered, and detection and isolation of pitch rate gyro faults affecting the mean and covariance matrix is addressed. A fault isolation technique based on the partition of s-dimensional innovation sequence into s one-dimensional innovation sequences, is presented, and the structure of a fault tolerant system is given.