Jafar Zarei

  PhD Thesis Defense Session

  Abstract:

  Fault diagnosis of complex control systems is one of the most important research topics among control engineering community during the last half century. In recent years, attention has tend ed towards designing of robust fault detection and diagnosis approaches both for linear and nonlinear systems. Many of the proposed methods in this field are based on robust observers which can efficiently estimate system states, in the presence of a wide class of modeling uncertainty and external disturbances. It should be noted that robust state estimation needs special conditions, which will limit its implementation. For nonlinear systems complexity is higher. One of the goals of this thesis is to present a robust method with good accuracy and simplicity compared to the so far presented methods for robust fault detection of nonlinear system.

  For this purpose, first, a method using the Unscented Kalman Filter (UKF) algorithm for detection of faults in nonlinear systems is proposed. Then convergence of the unscented Kalman filter is investigated and its local convergence conditions are obtained. To show the ability and performance of the presented method it is applied to detect and isolate faults of a nonlinear process, and it is shown that the presented method is superior to linearization-based methods such as Extended Kalman Filter (EKF). Since this filter lonely cannot solve the problem related to presence of uncertainty in the model, the concept of unknown input is introduced, and Unknown Input Observer (UIO) is used for fault detection of linearized model of the presented nonlinear system around the point of operation. It is shown that by considering an unknown input in the system and tacking into account its effect on the design of the observer, both unknown input effects and uncertainties can be decoupled from fault effects.

  UIO is then extended to a more general case and it will be shown that the algorithm presented by Kalman can be used to calculate the observer gain. Therefore, if the extended Kalman methods for nonlinear systems; such as EKF or UKF is used, UIO can be designed without linearization for fault detection purposes in nonlinear systems.

  Finally, convergence of the proposed observer will be studied and to demonstrate its ability it is applied to the introduced system. It is shown that the proposed observer is able to distinguish between fault effects and uncertainties or unknown inputs.

  Defense Date: Monday, February 14, 2011 Time: 17:00

  Place : Electrical Engineering Department, Room 305