I. Dumitrache

Genetic learning of fuzzy controllers

New methods for designing and analyzing fuzzy controllers are required. Some architectures for integrating genetic algorithms with fuzzy logic controllers, the so called hybrid geno-fuzzy controllers, are introduced and discussed. A new hybrid geno-fuzzy controller based on the algebraic model of the fuzzy controller is proposed. Genetic algorithms are shown to be able to deduce the algebraic model of a simple fuzzy controller used for controlling a truck backer–upper system. The genetic algorithm is further used to tune the coefficients of the deduced algebraic model. The simulated results indicate that the hybrid geno-fuzzy controller is superior to a conventional fuzzy controller.

Advanced adaptive techniques for multivariable nonlinear processes

Abstract Neural-network techniques are investigated in an application to the identification and subsequent on-line control of a process exhibiting nonlinearities and typical disturbances. The method proposed consists from a novel identification technique based on extended memory adaptation (EMA) and an efficient implementation of the predictive control based on a nonlinear programming method. An forced circulation evaporator was chosen as a realistic nonlinear case study for the techniques discussed in the paper.

A DES-supported agent-based control architecture for flexible manufacturing systems

Abstract The paper presents a methodology for the synthesis of a supervisor based on a Petri Net model. This supervisor is used for the real-time control of a Flexible Manufacturing System by a two layer agent-based architecture. The structure and functioning of the architecture as well as the FMS modeling approach are also described in the paper. The synthesis methodology was implemented into a software application, which was used on a case study.

Intelligent Supervisor for Manufacturing Systems

Abstract The paper presents a supervisory structure for manufacturing systems. Starting from the classic control theory in considering the attributions of the supervisor, the particular tasks for supervising a manufacturing system are listed. As a consequence, the structure of the supervisory system is deduced as being intelligent, hierarchic and distributed. Procedures for the supervisory task accomplishment are given.

Modified MLD Form for Discrete Optimization of Hybrid Systems

Abstract The paper investigates a new way of interconnecting the hybrid systems models with the classical predictive control philosophy. A modified MLD form is described which allows a discrete optimization approach for the predictive strategy. Performances in moving horizon control of hybrid systems can thus be improved by use of solving procedures for Quadratic (0,1) problems, in order to decrease the computation time. The transformation of continuous inputs is based on fuzzy intelligent techniques to increase the precision by adapting sets of variables

Adaptive Predictive Control Based on Neural Networks

Abstract Predictive techniques based on neural networks are investigated in an adaptive structure for on-line control of a process exhibiting nonlinearities and typical disturbances. The method proposed consists of a novel identification technique based on extended memory adaptation (BMA) and an efficient implementation of the predictive control based on a nonlinear programming method. A forced circulation evaporator was chosen as a realistic nonlinear case study for the techniques discussed in the paper.

Control of Time-Delay processes Based on Smith Predictors

Abstract One of the main problems of Smith predictor-based control of time-delay process is the availability of an adequate model. Industrial control algorithms usually operates with models with one or two time-constants and pure delay. The paper presents a control structure based on standard PI algorithm and off-line identification. Some implementation aspects are discussed, such as the influence of the sampling period and of the command limits, together with a robustness analysis.

Design patterns multiagents driven for software development

Abstract The complexity of nowadays systems imposes development of a dynamic architecture based on modularization, dependency, processing and data access principles. The field of software design and development is rich in patterns. Software patterns can be seen as a library of code modules that is reused during development of programs. The paper propose a software pattern multiagent systems driven. There will be taken four main patterns: singleton, transfer object, abstract factory and data access object and will be implemented as software agents. Patterns have always participants who are related through the pattern structure. This leads to the observation that different patterns can be created by varying type and features of the participants. An entire family of patterns can be built from one pattern; thus, there will be many smaller building blocks, cooperating and computing toward one goal. This is the trigger for using software agents as blocks. Each project functionality will be decomposed into many tasks, which will be completed by software agents. Each task will be solved by software agents acting as software pattern blocks. The agents (software pattern implementation) will be automatically generated based on the specificity of the task. The implementation is done using JADE, a mature software agent technology platform. The paper is focused on the automation of the task – software pattern(s) mapping and implementation.

USING MOBILE AGENT-BASED TECHNIQUES IN KNOWLEDGE MANAGEMENT FOR ORGANISATIONAL NETWORK INTEROPERABILITY

Abstract The paper will present an approach for the design of an interoperable knowledge management architecture that brings together concepts from agent-based control engineering and discrete event & hybrid modeling. The main goal of this architecture is to ensure knowledge interoperability for networks of organizations, transcending the ontological view.

ROBUST CONTROL OF NONLINEAR PROCESSES USING MULTIPLE MODELS

Abstract This paper addresses the problem of controlling a nonlinear process when linear models have been identified at different operating points. A multimodel approach is used to to identify and control highly nonlinear process. The internal linear model representation of the nonlinear plant for conventional robust controller is replaced by the linear parameter varying (LPV) model, which introduces transparency while offering distinct advantages for nonlinear model-based control. For this LPV structure a robust controller is then designed based on H ∞ techniques. Simulation results for a CSTR process are used to illustrate the performance benefits of multimodel approach.