Research in Systems and Control Theory at SZTAKI
by József Bokor
Systems and control theory are traditional research areas at SZTAKI. Recent topics include various aspects of system identification, uncertainty modelling and robust control, stochastic systems and control, fault detection and isolation with applications to process systems, vibration monitoring and diagnostics, fault tolerant and reconfigurable control systems.
Other topics are the integrated design of process and control system structures, modelling and control of cache systems, modelling and analysis of operating procedures by coloured Petri-nets. Applications of the results appear in a vision in the loop vehicle control system developed to avoid unintentional lane departure, or in the design and analysis of a new nuclear reactor protection system considered as a distributed hybrid computer control system. Most of these results are related to the activity of the Systems and Control Laboratory, in co-operation with the Combinatorial Computer Science Research Group and the Automation and Vehicle Engineering Departments of the Technical University Budapest and with the Stochastic Systems Research Group. Some of these topics will be characterized below.
Simultaneous Identification and Control
In model-based control design, the interaction between the controller and the identified model must be taken into account. Designing the controller, the difference between the model and the plant must be considered, and the control law will influence the choice of system identification approach. The aim of the iterative design is to improve the performance properties of the controlled system step by step. The main issue is to examine the interplay between modelling performance, control performance, and robustness, and to develop iterative schemes to solve different control engineering problems. Several closed loop identification methods are applied in the iterative schemes and it is investigated how the gray-box identification method, which takes the physical knowledge of the plant into account, fits to the iterative scheme. The cooperation between the worst-case identification process and the robust control design based on Hinf-norm is also investigated. Moreover, the iterative scheme for mixed H2/Hinf controller design that provides disturbance rejection and robustness against uncertainties is also developed.
Fault Detection, Isolation, Fault-tolerant and Reconfigurable Control
Fault detection and isolation consist of two basic steps: robust residual generation and residual evaluation. In our approach the problem of robust residual generation for the detection and isolation of faults by using state observers is addressed. Previous results have shown that if failure modes and unknown inputs appear in the independent subspaces of the state space then geometrical methods for decoupling their effects can be used in enhancing the robustness of the detection process. Utilising the geometric view and the Lie algebra structures associated to this filtering problem, detection filters for bilinear and more general classes of non-linear plants can be derived. If perfect decoupling is not possible, a robust detection filter design problem is set up as an Hinf filtering problem where the effect of estimation weighting on filtering performance is optimised through the input diagonal scaling of the estimation. By the proper choice of the estimation weight, the objective is to provide the smallest scaled L2 gain for the disturbance input. The problem of obtaining bounds on the scaled L2 gain of the system can be solved as a convex feasibility problem.
Integrated Vehicle Control System
In the past few years the vision-based sensing of the surrounding environment has gained a strategic importance not only in military applications but also in automotive field. In co-operation with Knorr-Bremse Brake Systems Ltd., the Systems and Control Laboratory develops an experimental vision-based driving assistance system for the automatic detection of the unintentional lane departure of road vehicles. The system is tested in various road, weather and lighting conditions. Detection is the first step towards a control system aimed at keeping the vehicle within the lane bounds. Intervention is achieved by using differential braking - the unequal application of braking torque to the left and right-side wheel. Adaptive and robust control strategies are investigated for the realisation of control of this brake system.
Integrated Design of Process and Control System Structures
Process and control system structures can be described by signed directed graph (SDG) models. Computationally efficient methods are developed for integrated (joint) design of these structures for various decentralized controller design strategies (eg LQR, disturbance rejection etc.).
Modelling and Control of Cache Systems
Cache systems are hierarchical distributed computer systems communicating with various protocols. Discrete dynamic models of existing cache systems are developed, investigated and verified using measured network data. The dynamics of various cache systems is described by discrete event and Petri-net models. Discrete control strategies are developed and studied in order to increase the capacity security and reliability of these systems.
Modelling and Identification of Signals and Systems in Rational Bases
Modelling signals and systems is a central issue in various problems of signal processing, change detection or model based control design. Signals can be described eg as elements of abstract function spaces like L2 or l2, and systems are considered as operators mapping an input signal space to an output one. A usual representation of signals in Hilbert spaces is given by defining a shift operator and to construct an orthogonal decomposition. The signals are represented as infinite sums where the coefficients are obtained by projections. The well known engineering representation is related to the canonical shift on the unit elements of the space l2. Using Z-transforms, this leads to an IIR representation in polynomial basis (the powers of the complex variable z) in the Hardy space H2. FDLTI systems mapping this signal space to an output H2 space can be represented by rational functions.
The role of choosing the shift operator (or alternatively to choose a basis) in the signal spaces and to investigate its effect on the properties of input/output operators associated to these bases opened a vigorous research some years ago. A very active and inspiring role has been played by the Delft team of P.M.J. van den Hof moreover by B. Walberg and B. Ninness and coworkers, see their Preconference Workshop on the IFAC Congress in Beijing. The Systems and Control Laboratory joined this research by investigating the construction of uniformly bounded operators associated to the Blaschke shift on H2. The form of these operators has been extended to approximate operators in the Hinf norm, too. These results can be used to identify models either under H2 or Hinf norm criteria when using rational bases instead of the polynomial one. This approach has a number of advantages when solving a large family of detection and identification problems. A generalization of the Ho-Kalman realization method has been elaborated too, to provide state space models for control system design.
References can be found via the home page of SZTAKI at http://www.sztaki.hu/
József Bokor - SZTAKI
Tel: +36 1 209 6990