2000 (Volume 10)
Some Remarks on Adaptive Stabilization of Infinite-Dimensional Second Order Systems
(Kyushu Institute of Technology, Japan)
In this paper adaptive stabilization of infinite-dimensional undamped second order systems is considered in the case of the input and output operators being collocated. The advantage of the adaptive law is that good control performance can be automatically achieved even in the presence of various types of uncertainties. The adaptive stabilizer is constructed by the concept of high-gain output feedback. An energy-like function and a multiplier function are introduced and adaptive stabilization of the linear second order systems is analyzed. It is shown that the theory of adaptive stabilization can be also applied to some nonlinear systems.
keywords: adaptive stabilization, undamped second order dynamical systems, energy-like functions, multiplier functions.
The Grafpol programming language for programmable logic controllers
|T. Mikulczyński, Z. Samsonowicz and R. Więcławek|
(Wrocław University of Technology, Poland)
In the paper, the new Grafpol programming language for PLCs is presented. The theoretical framework for the new language and its network models have been provided: the operation network and the Grafpol network. The syntax of the Grafpol language has been defined and described. Principles of notation of sequential and concurrent procedures used in algorithms for discrete manufacturing processes by means of the Grafpol language have been developed and presented. Compliance of the Grafpol language with requirements of the IEC:1131-3 standard referring to PLC programming languages has been proved.
keywords: programmable logic controller, programming language, discrete manufacturing processes.
Identification of coefficients of a low order continuous time transfer function from discrete-time recorded measurements
(Warsaw Technical University, Poland)
The paper deals with a problem of identification parameters of a continuous time (CT) transfer function for a low order linear systems with discrete-time (DT) recorded data. Algorithms for a direct estimation of the CT-coefficients are developed from rules for transformation of a CT-transfer function controlled via zero-order sampling unit into DT-representation. Two schemes are derived and tested: one based on the Goodwin transformation and second derived from the modified Tustin transformation. Both approaches have resulted in similar relations, which can be used for direct estimation of the CT-coefficients of the model of an investigated system. The numerical schemes contain some expressions, that are alike DT-differences and in effect they can magnify impacts of different disturbances. Therefore are presented results of extended testing of both schemes including different types disturbances; measurement noises, slow varying drifts, measurement resolution errors together with changes of the sampling time. As an object was used a model of a third order linear servomechanism system with oscillating and integration actions. A comparison with results determined by LS-recursive scheme are presented too.
keywords: identification, continuous time models, linear systems, discrete time data representation.
Convergence of a dual algorithm for minimax problems
(Dalian University of Technology, China)
(Chinese Academy of Science, China)
This paper studies the convergence of a dual algorithm for solving minimax problems proposed by Zhang and Tang (1997), which is based on a penalty function of Bertsekas (1982). It proves that the dual algorithm is locally convergent with linear convergence rate under the commonly used assumptions. Numerical results are presented to show the effectiveness of this algorithm.
keywords: minimax problem, penalty function, dual algorithm, convergence.
Neuro-fuzzy classifying system for intelligent decision support. Part I. Methodology Part II. Applications
|Marian B. Gorzałczany|
(Kielce University of Technology, Poland)
The description of complex decision making processes is usually based on the combination of two types of knowledge and data: a qualitative, fuzzy one which contains elements of uncertainty and vagueness and often is expressed in the form of linguistic rules usually provided by a domain expert, and a quantitative, non-fuzzy one which appears in the form of measurements and other numerical data. This paper presents a methodology for the design of decision support systems. This methodology can effectively learn, represent, process and generalize both qualitative and quantitative knowledge and data contributing to the description of complex decision making processes. The proposed approach combines artificial neural networks with the theory of fuzzy sets giving a structure that can be called a neuro-fuzzy classifier. Part I of this paper presents this classifier in both learning and approximate-inference phases. Two decision support systems designed with the use of the proposed neuro-fuzzy classifiers are presented in Part II of this paper.
keywords: fuzzy sets, artificial neural networks, neuro-fuzzy systems, neuro-fuzzy classifiers, decision support systems, expert systems.
Mathematical and Neural Network Modelling of a Wastewater Treatment Plant
|Lucyna Bogdan, Jan Studziński and Zbigniew Nahorski|
(Polish Academy of Sciences, Poland)
(Wrocław Agricultural University, Poland)
|Ryszard Szetela |
(Wrocław Technical University, Poland)
Comparison of few methodologies of building models useful in wastewater treatment plant maintenance is performed. One is mathematical modelling of the activated sludge process. It consists of modellling of the basic vessels: primary clarifiers, aerator basins and secondary clarifiers, linked and partially looped, as well as equations describing the physical and biochemical transformations going on in the vessels: sedimentation in the clarifiers and biological processes changing the influent wastewater chemical composition. The models' parameters were estimated in two steps. In the first step the active volumes of the vessels were estimated from the experiment performed in the plant. In the second step, parameters known from the literature were used as the initial guess and then calibrated to fit the observations taken during normal plant operation.
Concerning other methodologies, results from the black box modelling of the performance of the plant with the neural network are given. The neural network and the time series models are also applied for prediction of the influent wastewater.
keywords: wastewater treatment, biochemical processes, modelling, parameter estimation, neural networks.