Diz 82 en
State estimation and fault detection with reduced error sensitivity to paramaters
Author: Jaroslav Tabaček
This thesis extends the theory of desensitized filtering. Desensitized filtering is an efficient approach to state estimation for systems with uncertain parameters. The stochastic approach to sensitivity reduction developed in this thesis leads to the exact desensitized Kalman filter (XDKF) without using assumptions that are not fully justified. Based on this result, several variations of the XDKF are introduced for specific purposes. The stochastic approach allows providing a straightforward way to conduct stability analysis. The XDKF is also used with the interacting multiple model method, which results in fault detection and diagnosis (FDD) algorithms that work with simplified models. Another useful application of the XDKF was found in the distributed state estimation algorithms, where it helps to improve local estimation by considering neighbor estimate uncertainty without increased communication burden. This distributed approach is also used for developing a distributed FDD method, which can detect and diagnose local and global faults.