Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.504727
Title: Observer-based parameterestimation and fault detection
Author: Dai, Xuewu
Awarding Body: The University of Manchester
Current Institution: University of Manchester
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
Abstract:
This PhD work is motivated by on-board condition monitoring of gas turbine engines (GTEs) and presents a constructive robust fault detection procedure integrating system identification, time delay compensation, eigenstructure assignment, zero assignment and dynamic observer design techniques, to detect faults in a dynamic system corrupted by disturbances at some frequencies. The main results achieved in this PhD study are: (1) Application of nonlinear least squares to Output Error (OE) model identification. Although OE model shows better performance on long-term prediction, the challenge is the dependency within the long-term prediction errors. The dependency is tackled by an iterative calculation of the gradient, and an approximation of the Hessian matrix is adopted to accelerate the convergence. (2) Delay compensation for high-gain observer based time-varying parameter estimation. In the high-gain observer based parameter estimation, it is usually assumed that the estimation delay is zero. This assumption puts some constraints on the observer design and may not be satisfied in some situations. By examining the transfer function matrices associated with the high-gain observer, a novel time delay calculation and compensation approach is proposed. The main contribution is the proof of the fact that the estimation delay is free from the plant parameter variation. Then a nonlinear phase delay filter approximation technique is used to compensate the delay. (3) Zero assignment in dynamic fault detection observer design. It is well known in filter design that zeros have the ability to block the propagation of some input signal through the system at some frequency. In this thesis. this idea is used to assign zeros to the desired places so that the disturbance can be attenuated. In most observer design research, however, the structure is confined to the classic (static) Luenberger structure where the gain is a constant. numerical matrix. As proved in this thesis, zeros of static observers arc invariant. Hence the dynamic observer is proposed, where a dynamic system (dynamic feedback gain) substitutes for the constant numerical gain matrix. As a result, some additional zeros are introduced and can be assigned arbitrarily to the desired places. To the best of our knowledge, although the concept of zeros in multivariable systems has been proposed by Rosenbrock over thirty years, there have been no known results of utilising zero assignment to robust fault detection observer design.
Supervisor: Not available Sponsor: Not available
Qualification Name: Not available Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.504727  DOI: Not available
Share: