Listagem por temas:
Escolhendo um título você terá acesso ao arquivo original em Post-Script.
COB169 INTELIGÊNCIA COMPUTACIONAL APLICADA À MODELAGEM DE UM TURBO-GERADOR: ABORDAGEM NEURAL E EVOLUTIVA / COMPUTATIONAL INTELLIGENCE APPLIED FOR MODELLING OF A TURBOGENERATOR: NEURAL AND EVOLUTION APPROACHES
Leandro dos Santos Coelho & Antonio Augusto Rodrigues Coelho
Departamento de Automação e Sistemas, UFSC, Caixa Postal 476
CEP 88040.900 Florianópolis, Brasil -E-mail: {lscoelho; aarc}@lcmi.ufsc.br
This paper evaluates the application of computational intelligence methodologies in a nonlinear process identification. The different intelligent methodologies are evolutionary computation and artificial neural networks (feedforward and recurrent topologies). The simulations are realized in the identification of a turbogenerator mathematical model through a step signal, pseudo-random binary sequence, and white noise excitation signals. The performance of the techniques are presented and discussed.
Keywords: Process Identification, Evolutionary Computation, Genetic Algorithms, Evolution Strategies, Artificial Neural Networks / Identificação de Processos, Computação Evolucionária, Algoritmos Genéticos, Estratégias Evolutivas, Redes Neurais Artificiais.
COB170 CONTROLADORES NEBULOSO E NEURAL COM OTIMIZAÇÃO EVOLUTIVA: METODOLOGIAS E APLICAÇÃO / FUZZY AND NEURAL CONTROLLERS WITH EVOLUTION OPTIMIZATION: METHODOLOGIES AND APPLICATION
Leandro dos Santos Coelho & Antonio Augusto Rodrigues Coelho
Departamento de Automação e Sistemas, UFSC, Caixa Postal 476 - CEP 88040.900 Florianópolis, Brasil
E-mail: {lscoelho; aarc}@lcmi.ufsc.br
This paper presents, evaluates and compares different intelligent strategies in process control. The design and configuration of the controllers is realized by the following hydrid methodologies: i) fuzzy control with evolutionary optimization of membership functions and, ii) control via Elman recurrent topology with evolutionary training. The evolutionary hybridization of the controllers is realized through the optimization and tuning, which in turn are set by, evolution strategies with self-adaptation mechanisms. Experimental tests are conducted to analyse the control techniques for dealing with a nonlinear system in a liquid-level regulation submitted to reference and load disturbances.
Keywords: Intelligent Control, Evolution Strategies, Fuzzy Control, Neurocontrol, Level plant / Controle Inteligente, Estratégias Evolutivas, Controle Nebuloso, Controle Neural, Planta de Nível.
COB282 CAVITATION DETECTION IN HYDROTURBINES USING NEURAL NETWORKS
Shih Man Lin, João Souza Neto, Antonio C. P. Brasil & Danilo Santos
Univ. de Brasília/FT/Dpto. Eng. Mecânica, Campus Universitário, Asa Norte, CEP 70910-900 Brasilia/DF. Email:manlin@enm.unb.br
This paper refers to the development of a methodology to detect cavitation in hydroturbines using a neural network strategy. Both experimental tests and theoretical analysis were carried out. Initially, the methodology of mapping the cavitation regimes in Francis turbines was proposed and tested. The neural network reproduced satisfactorily the different types of cavitation. A real scale experiment was also performed in a 160 MW Francis Turbine in operation at the electric power station in Ilha Solteira/São Paulo (CESP). Acoustic sensors were used to perform preliminary tests on cavitation radiated noise, in order to detect cavitation for different conditions of turbine operations. The neural network methodology was also proposed to analyze these experimental data. A description of laboratory facilities and some results obtained up to this moment can also be found in this paper.
Keywords: Cavitation, TurboMachinery, Neural Networks and Neuro-Genetic Systems/Cavitação, Máquinas Hidráulicas, redes neurais e sistemas neuro-genéticos
COB410 REDES NEURAIS APLICADAS AO CONTROLE DE ATITUDE DE SATÉLITES ARTIFICIAIS COM APÊNDICES FLEXÍVEIS/NEURAL NETWORK APPLIED TO ATTITUDE CONTROL FOR ARTIFICIAL SATELLITE WITH FLEXIBLE APPENDAGES.
Sebastião E. C. Varotto & Atair Rios Neto
Instituto Nacional de Pesquisas Espaciais – INPE/MCT,
CEP 12201-970 CP 515 São José dos Campos, SP. E-mail: varotto@dem.inpe.br
Instituto de Pesquisa e Desenvolvimento, Universidade do Vale do Paraíba.
CEP 12245-720 São José dos Campos, SP. E-mail: atair@univap.br
This work demonstrates that artificial neural networks can be used effectively for satellite attitude dynamics identification and control. In order to exemplify this application, a satellite with a rigid main body, three reaction wheel and three flexible solar panel was chosen (lay-out similar to Brazilian Remote Sensing Satellite) The main objectives of this work are to test the neural control and analyze the interaction between the control system and the elastic motion of the satellite solar arrays. The equations of motion were derived by the Lagragian approach for quasi-coordinates (rotational motion) and for generalized coordinates (elastic motion).The identification of neural nets parameters is performed by Kalman filtering algorithm with a local parallel processing version..
Keywords: Redes neurais artificiais, sistemas não lineares, filtro de Kaman, controle de atitude. Artificial neural networks, non-linear system, Kalman filtering, satellite attitude control.
COB487 Detection of state-space parameter perturbation using recursive nonlinear estimators
Belisário Nina Huallpa(*), Eurípedes Nóbrega(*) & Fernando José Von Zuben(**)
(*)DMC-FEM-UNICAMP – Brazil - Caixa Postal 6122 – CEP: 13083-970 - email: beli @dmc.fem.unicamp.br, egon@dmc.fem.unicamp.br
(**)DCA-FEEC-UNICAMP - Brazil - Caixa Postal 6101 – CEP: 13083-970 - email: vonzuben @dca.fee.unicamp.br
The problems of parameter estimation and perturbation detection are central to dynamic system identification. The purpose of this paper is to present a robust numerical tool designed to detect permanent parameter perturbation on nonstationary dynamic systems. The main component of the detection apparatus is a Hopfield neural network operating as a recursive nonlinear parameter estimator. At the first step, the Hopfield neural network is applied to estimate the initial parameter values in the state-space model of the dynamic system. The resulting state-space parameter obtained after convergence are assumed to be the nominal parameter for that application. It is necessary to let the estimation process converge to the nominal values before starting the detection of parameter perturbation. So, after convergence, sensitivity analysis is recursively applied to promptly detect any kind of significant perturbation in each output of the Hopfield neural network, with transient perturbations being discarded. The detection process must be robust and flexible enough to deal with a large range of dynamic behavior, because the shape and magnitude of the permanent perturbation, together with its effect on the dynamic system, cannot be properly anticipated.
Keywords: Hopfield neural network, state-space models, parameter estimation, permanent perturbation detection
Redes de Hopfield, modelos de espaço de estados, estimação de parâmetros, detecção de perturbação permanente
COB643 MODELAGEM DO PROCESSO DE USINAGEM POR FRESAMENTO UTILIZANDO REDES NEURAIS / MODELLING MILLING MACHINING PROCESS USING NEURAL NETWORK PROCEDURE
André L. B. Dos santos, Marcus A. V. Duarte e Carlos R. Ribeiro
Departamento de Engenharia Mecânica - Universidade Federal de Uberlândia
CEP 38.400 - 089 Campus Sta. Mônica, MG, Brasil. E-mail: mavduarte@ufu.br
Usually Taylor’s equation is one of the most used models to represent the tool life in machining processes. Unfortunately, the great deal of cutting tools available as well as the large range of cutting conditions together with the complexity of some machining processes, all make the reliability of the results obtained via this equation limited. This work presents a neural network procedure in order to estimate the cutting tool life for milling operation. ABNT 1045 steel bars and triple coated cemented carbide tools were used in order to validate the proposed methodology. The results indicate that the suggested neural network procedure povide a considerable reduction in the erro when predicting the machining time and when compared with those obtained from optimization procedures.
Keywords: Equação de Taylor, Vida da Ferramenta de Corte, Fresamento, Modelagem Utilizando Redes Neurais, Experimento Ótimo / Taylor’s Equation, Cutting Tool Life, Milling, Modelling, Neural Network, Optimal Experiment.
COB687 IDENTIFICAÇÃO DE MODELOS DINÂMICOS DE SATÉLITES COM GEOMETRIA VARIÁVEL ATRAVÉS DE REDES NEURAIS / VARIABLE GEOMETRY SATELLITE DYNAMICS IDENTIFICATION WITH NEURAL NETWORKS
Valdemir Carrara & Atair Rios Neto
The use of neural networks for satellite attitude dynamics identification is addressed in this work. In order to validate this application, a spacecraft with a variable dynamic behavior due to articulated appendages fixed to the body was chosen. The differential equations therefore show the nonlinear dynamic effects to be identified by neural nets. In this work some of the main expressions that allow system modeling through neural nets as well as a least squares based training procedure are presented. A general method for obtaining the inertia tensor and center of mass of an articulated space device, is also explained together with the dynamic and cinematic differential equations. These formulations were used in attitude simulation for neural network system identification and control training’s.
Keywords: Neural networks, satellite attitude, non-linear dynamics, system identification. Redes neurais, atitude de satélites, dinâmica não-linear, identificação de sistemas.