Threshold selector for fault detection on closed-loop predictor-based recursive system identification

Abstract

In this research, a technique is developed to separate an additive fault from a multiplicative fault by using recursive update of predictor-based system identification. If additive fault occurs at the same moment of multiplicative fault in a system, the recursively updated system model can take away the effect of multiplicative fault from residual. Thus, purely additive fault affects on residual of which threshold is calculated based on Chi-squared distribution. In this research, the Chi-squared residual is computed as whitened-residual and its whitening covariance matrix is derived from residual itself in order to accommodate identification error. To detect multiplicative fault, the norm to measure the distance between the original system matrices and the changed is recursively checked with user-defined parameter. The benefit of this approach is illustrated with an adopted system.