Kalman Filtering for Uncertain

To my family, anna and ammu ACKNOWLEDGEMENT I would like to express my sincere indebtness and gratitude to my thesis advisor Dr. Dan Simon, for the ingenious commitment, encouragement and highly valuable advice he provided me over the entire course of this thesis. I would also like to thank my committee members Dr. Zhiqiang Gao and Dr. Sridhar Ungarala for their support and advice. I wish thank my lab mates at the Embedded Control Systems Research Laboratory for their encouragement and intellectual input during the entire course of this thesis without which this work wouldn't have been possible. Finally I would like to mention special thanks to Mr. Don Simon at NASA, GRC whose support made thesis a reality. ABSTRACT Aircraft health monitoring has been a challenging task for over decades. In turbofan jet engines all the parameters which describe the health of the engine cannot be measured explicitly. One possible solution to this problem is Kalman filter. The traditional Kalman filter is optimal as long as the modeling of the plant is accurate. The turbofan jet engine being highly non-linear makes the task difficult. This thesis shows a way of linearizing the jet engine model so that theoretically proven estimation techniques can be applied to this problem. This thesis presents the application of Kalman filter to health parameter monitoring of the gas turbine engine. It is shown that the standard Kalman filter will not be robust enough if there are uncertainties in the modeling of the plant. A new filter is developed in this thesis which addresses the uncertainties in the process noise and measurement noise covariances without assuming any bounds on them. A hybrid gradient descent algorithm is proposed to tune the new filter gain. This filter is then implemented for the health parameter estimation. The results show significant decrease in the estimation error covariance. It is shown in the conclusions that advanced search algorithms like Genetic Algorithms proves to be superior to hybrid gradient descent algorithm in searching for better minima.