Fault isolation with estimated frequency response

Abstract

In this paper, a frequency domain fault isolation method for linear time-invariant system is proposed. A fault is assumed to cause a change in one of the parameters of the transfer function of the plant, leading to a change in the frequency response. Based on this changed frequency response, a fault isolator is designed to determine which parameter has changed. This is achieved without any estimation of the parameter value. The proposed method captures the changes in the frequency response by an observation vector involving estimates of the frequency response. The decision on which parameter has changed is based on the generalized likelihood ratio test. The performance of this isolator is defined to be the probability of correct isolation and it can be calculated easily using the idea of 'correct' spaces. Simulations are given to illustrate the results.