A nonlinear version of the generalized likelihood ratio test

In this paper, a new version of the generalized likelihood ratio (GLR) test is proposed to deal with the fault diagnosis problem in nonlinear systems. The unscented Kalman filter (UKF) is utilized for state estimation and innovation generation. The fault signatures used in the hypotheses test are pre-computed by running the fault model and the nominal model simultaneously based on the state estimations. In such a way the intolerable computational problem, which arise from applying a bank of nonlinear filters, is solved. Further more, the fault isolation problem as well as the fault estimation problem of the non-abrupt fault is discussed based on the nonlinear GLR approach. The simulation study of a longitudinal aircraft model is utilized to show the effectiveness of the proposed method.