Affordable Access

Robust observer design for nonlinear systems

  • Gonzalez de Cossio, Francisco
Publication Date
Dec 05, 2019
Kaleidoscope Open Archive
External links


Estimating the state of a nonlinear system is an essential task for achieving important objectives such as: process monitoring, identification and control. Observers are algorithms that estimate the current state by using, among other information, sensor measurements. The problem of observer design for nonlinear systems has been a major research topic in control for many decades. Recently, there has been an increasing interest in the design of observers for more realistic models, which can include disturbances, sensor nonlinearities and discrete outputs. This thesis concerns the design of robust observers for selected classes of nonlinear systems and we can distinguish three main parts. The first part studies state-affine systems affected by noise, and analyses the state estimation via the so-called high-gain Kalman filter. The convergence properties of this observer are strongly influenced by two variables: its tuning parameter and the properly excited system input. We present a new optimization algorithm, based on Lyapunov analyses, that adapts these variables in order to minimize the effect of both dynamic and output disturbances. The novelty of this approach is that it provides a systematic method of simultaneous tuning and input selection with the goal of improving state estimation in the face of disturbances, and that it avoids the use of trial-and-error based methods. The second part studies the problem of observer redesign for general nonlinear systems whose outputs are transformed by nonlinear functions. Indeed, a given observer might not estimate the system state properly if it does not take into account sensor nonlinearities and, therefore, such an output mismatch needs to be addressed. We present an observer redesign that consists in the interconnection of the original observer with an output estimator based on a dynamic inversion, and we show its asymptotic convergence via small-gain arguments. We illustrate our method with two important classes of systems: state-affine systems up to output injection and systems with additive triangular nonlinearity. Finally, the third part extends our redesign method to systems whose outputs are not only transformed but also discretized in time. This added assumption introduces important challenges; we now implement sample-and-hold techniques leading to an observer gain based on linear matrix inequalities. The main feature of our redesign methods is the possibility to adapt a large number of observers from the literature to more realistic scenarios. Indeed, classical sensors in engineering applications are often nonlinear or discrete, whereas a recurrent assumption in observer design is the linearity or continuity of the output

Report this publication


Seen <100 times