Witold Byrski

Context-aware reconfiguration of autonomic managers in real-time control applications

We consider autonomic applications to systems for which continuous perfect monitoring of state is not possible. We use Exact-State Observers (ESO) to provide enhanced information about the system state. To achieve optimal configuration of the autonomic controller itself, over a wide range of environmental operating conditions, and across a wide range of unique application domains, we implement a new architecture for dynamic supervision and control systems in which a policy-based autonomic engine automatically selects both its monitoring and actuator components to suit ambient operating conditions. By using a suite of ESOs tuned for different tradeoffs between real-time responsiveness and extent of system disturbance tolerated, and a policy mechanism to contextually select the most appropriate observer at any given time, we achieve self-configuring and self-optimising behaviours whilst keeping the complexity, resource-requirements and adaptation latency low.

The role of parameter constraints in EE and OE methods for optimal identification of continuous LTI models

The role of parameter constraints in EE and OE methods for optimal identification of continuous LTI models The paper presents two methods used for the identification of Continuous-time Linear Time Invariant (CLTI) systems. In both methods the idea of using modulating functions and a convolution filter is exploited. It enables the proper transformation of a differential equation to an algebraic equation with the same parameters. Possible different normalizations of the model are strictly connected with different parameter constraints which have to be assumed for the nontrivial solution of the optimal identification problem. Different parameter constraints result in different quality of identification. A thorough discussion on the role of parameter constraints in the optimality of system identification is included. For time continuous systems, the Equation Error Method (EEM) is compared with the continuous version of the Output Error Method (OEM), which appears as a special sub-case of the EEM.

Optimal Identification of Continuous Systems and A New Fast Algorithm for On-Line Mode

Abstract The optimal method for identification of parameters Θ of continuous linear SISO system by the use of modulating functions is presented. Solution to this problem leads directly to minimization of quadratic form ΘTGΘ on the unit ball ΘTΘ=1 where G obtained from measurements is real, symmetric, positively defined Gram matrix and vector Θ∈Rn is unknown and should be found. The optimal solution for this minimization is eigenvector W1 related to minimal eigenvalue λ1 of matrix G. The main new problem stated and solved in this paper is the fast algorithm based on the behavior of some nonlinear differential equation which is fulfilled on eigenvector w1(t)=Θ(t) of nonstationary matrix G(t) obtained in on-line mode.

A Double Window State Observer for Detection and Isolation of Abrupt Changes in Parameters

Abstract The paper presents a new method for diagnosis of a process fault which takes the form of an abrupt change in some real parameter of a time-continuous linear system. The abrupt fault in the process real parameter is reflected in step changes in many parameters of the input/output model as well as in step changes in canonical state variables of the system. Detection of these state changes will enable localization of the faulty parameter in the system. For detecting state changes, a special type of exact state observer will be used. The canonical state will be represented by the derivatives of the measured output signal. Hence the exact state observer will play the role of virtual sensors for reconstruction of the derivatives of the output signal. For designing the exact state observer, the model parameters before and after the moment of fault occurrence must be known. To this end, a special identification method with modulating functions will be used. A novel concept presented in this paper concerns the structure of the observer. It will take the form of a double moving window observer which consists of two signal processing windows, each of width T. These windows are coupled to each other with a common edge. The right-hand side edge of the left-side moving window in the interval [t − 2T, t − T ] is connected to the left-hand side edge of the right-side window which operates in the interval [t − T, t]. The double observer uses different measurements of input/output signals in both the windows, and for each current time t simultaneously reconstructs two values of the state—the final value of the state in the left-side window zT (t − T) and the initial value of the state z0(t − T) in the right-side window. If the process parameters are constant, the values of both the states on the common joint edge are the same. If an abrupt change (fault) in some parameter at the moment tA = t − T occurs in the system, then step changes in some variables of the canonical state vector will also occur and the difference between the states will be detected. This will enable localization of the faulty parameter in the system.

On-line fast identification method and exact state observer for adaptive control of continuous system

Paper presents on-line cooperation of two advanced signal processing methods which are used for the best adaptation of LQ controller during stabilization of continuous linear system CLS in which abrupt changes of parameters have occurred. First processing algorithm is connected with special modification of modulating functions method MFM operating in two moving windows for fast identification of step changes in parameters. The second processing algorithm is used for exact state observation by the use of optimal integral observer operating in the third moving window. The modulating functions method with fixed moving convolution window is known for identification of time-invariant system CLTIS. The application of MFM to the problem of identification of step changes in the parameters causes the appearance of fixed time delay in the reconstruction of the new values of parameters after their failure. The aim of this paper is presentation of the concept of two different modifications of MFM which enable fast optimal identification of parameter faults with minimization of time delay as well as general idea for LQ adaptive controller.

Design and Implementation of a New Algorithm for Fast Diagnosis of Step Changes in Parameters of Continuous Systems

Abstract In this paper special adaptation of modulating functions method MFM for fast detection of step changes in parameters of continuous and linear systems is presented. The modulating functions method is well known for identification of continuous, linear, time-invariant system CLTIS. The characteristic feature of MF method in its application to the problem of identification of step changes in parameters is the existence of fixed time delay in the reconstruction of the new value of parameters after their failure. The aim of this paper is presentation of the substantial modification of MFM which enables optimal identification of parameter faults with minimization of time delay.

Context-aware real-time systems with autonomie controllers

This paper considers autonomie applications to systems for which continuous perfect monitoring of state is not possible. We use sophisticated exact-state observers to provide enhanced information about the system state. In order that we can achieve optimal configuration of the autonomie controller itself over a wide range of environmental operating conditions, and across a wide range of unique application domains, we implement a new architecture for dynamic supervision and control systems in which a policy-based autonomie engine automatically selects both its monitoring component (a state observer) and its actuator component to suit ambient operating conditions. We show how the hybrid approach enables autonomie computing to be deployed in systems in which the target system cannot be monitored continually and perfectly. We also show how the observer window size can be dynamically selected to give the best controller performance for any level of system disturbance.

Integral Observer of Finite State in Heat Equation Approximated by ODE

Abstract In the paper the perfect reconstruction of the state vector of the large dimension linear system from the measurement of input and output data by the use of the observer is presented. The structure of the proposed observer originates directly from the definition of the exact observability hence it is based on the integral and not on the differential equation. In previous works the general approach to exact observation of finite state vector in linear time invariant system by the use of integral deterministic observers was stated. The problem of optimal state reconstruction under assumption that errors in output and control measurements occur was also solved. These results are applied to distributed parameter model approximated by the large dimension set of ordinary differential equations (ODE).

Laboratory Distillation Column for Education on Process Control

Abstract A laboratory rig aimed at education on industrial computer control systems is presented. The laboratory installation consists of a small scale glass distillation column and related technological devices (evaporator, heat exchangers, tanks, pumps), equipped with measurement system (63 sensors) and control valves, connected to industrial programmable digital controllers. The controllers perform primary control tasks. More advanced algorithms (monitoring, MIMO control) are carried out at PC compatible industrial computers working in a network structure (linked with the primary controllers). The laboratory enables demonstration of different control problems, starting from the simple digital control design up to implementation of newest MIMO control techniques.

LQ Control and EKF as the State Observer in Stabilization of Distillation Column

Abstract In the paper an application of the LQ control strategy and Extended Kalman Filter to the stabilization of the profile of tray compositions in the distillation column is considered. The multivariable model of the plant is given by the set of nonlinear differential equations which follow from material balances and equilibrium equations. Numerical results of simulation experiments and their discussion are given