Study On Adaptive Observer Design In A Constructive Manner And Its Applications Fault Diagnosis

Joint estimation of states and parameters in state-space systems is of practical importance for fault diagnosis and for adaptive control. Recursive algorithms designed for this purpose are usually known as adaptive observers. In this dissertation a class of adaptive observers in a constructive manner is discussed and the main contributions are described as follows:1. Previous works on globally convergent adaptive observers do not consider unknown parameters in both state equations and output equations. In this dissertation for continuous-time multiple-input multiple-output linear time-varying systems with unknown parameters in both state and output equations, an adaptive observer with exponential forgetting factor is designed in a constructive manner and its global exponential convergence is formally established under appropriate assumptions. Compared to the Kalman filter, the advantages of the proposed adaptive observer reside both in its simpler convergence condition and in its more efficient numerical implementation. A numerical example is presented to illustrate the performance of the adaptive observer.2. For discrete-time multiple-input multiple-output linear time-varying systems with unknown parameters in both state and output equations, an adaptive observer is designed in a constructive manner. In order to establish the global exponential convergence for simultaneous estimation of states and unknown parameters, a persistent excitation condition is required. A numerical example is presented to illustrate the performance of the adaptive observer.3. For a class of single output uniformly observable nonlinear systems with unknown parameters in both state and output equations, an adaptive observer is designed in a constructive manner based on the techniques of high gain observer and adaptive estimation. The high gain adaptive observer for joint state and unknown parameter estimation is conceptually simple. The new algorithm makes use of a time varying gain matrix for unknown parameter estimation, which simplifies the initialization and parameter tuning. In order to establish the global exponential convergence of the adaptive observer, a persistent excitation condition is required. Consequently, the global exponential convergence for simultaneous estimation of states and unknown parameters is formally established following a simple procedure. A numerical example is presented to illustrate the performance of this adaptive observer.4. Furthermore a more general adaptive observer for nonlinear multi-input-multi-output (MIMO) systems is developed and a persistent excitation condition is obtained in order to establish the global exponential convergence of the adaptive observer. A numerical example is presented to illustrate the performance of this adaptive observer.5. A residual is a transformation of raw data with some desired properties for the purpose of fault diagnosis. Typically, model-based methods for fault diagnosis rely on the design of residuals which are signals computed from available sensor measurements with the aid of a mathematical model of the monitored system. Based on the developed adaptive observer in this dissertation, residuals are generated for fault diagnosis in linear time-varying systems. The sensitivity of the residuals to the monitored faults is rigorously analyzed, as well as their insensitivity to the faults to be ignored. The residuals presented in this dissertation extend the results of previous works by simultaneously taking into account both actuator and sensor faults. A numerical example is presented to illustrate the performance of these residuals.6. A full vehicle active suspension system with the nolinear dynamics is considered. The aging coefficients of the components are modeled as unknown parameters. An adaptive observer is designed to estimate the aging coefficients. The simulation result shows the aging coefficients could be estimated rapidly.