Abstract
We have proposed a robust fault detection approach for nonlinear systems. In the method, Kullback Discrimination Information(KDI) is introduced as a fault detection index to discriminate model distortion due to a fault. Then fault detection is executed based on threshold approach. However it has been difficult how to determine appropriate threshold value η from a statistical view point. Therefore we adopt a method for setting η based on the distribution of the KDI learned by using a potential function approach. Through a numerical example, we verify the effectiveness of the method.
