Abstract: A fault diagnosis approach based on least squares estimation(FDBLSE) for a class of nonlinear systems is proposed in this paper. It first constructs a state estimator for the systems to be diagnosed so that the dynamic relationship between state estimation error and faults is obtained, it then identifies the faults using least squares estimation method based on the obtained dynamic relationship. The fault identification error and the robustness, sensitivity to faults and the detection time of the fault diagnosis are analyzed. A deep-going comparison between the approach proposed in this paper and the fault diagnosis approach based on the learning method is also given. FDBLSE approach can not only identify faults, but also give the upper limit of the identification error. It highlights short identification-time and high accuracy of fault identification. The simulation indicates the effectiveness of the approach proposed in this paper.

Key words: least squares estimation, sensitivi-ty, fault diagnosis, detection time, fault identification, robustness