Inteligência computacional aplicada à adaptação na modelagem matemática do processo de laminação a frio de aços planos.

Abstract

Inteligência Computacional (mais especificamente, Redes Neurais Artificiais) para tornar mais eficiente a adaptação de dois importantes parâmetros: o limite de escoamento do material e o coeficiente de atrito entre a tira e os cilindros de trabalho. O texto apresenta os fundamentos teóricos, a metodologia de desenvolvimento e a implantação da solução, bem como os resultados preliminares obtidos com a aplicação da proposta num laminador real. Finalmente, discutem-se esses resultados, apresentam-se as conclusões, sumarizam-se as contribuições da pesquisa e se sugerem futuras linhas de pesquisa para extensão dos resultados. Palavras-chave: Automação industrial. Laminação. Redes neurais. Inteligência artificial. ABSTRACT SANTOS FILHO, A. L. Computational intelligence applied to the adaptation of mathematical modeling in flat steel cold rolling process. 2009. 179 p. Tese (Doutorado) -Escola Politécnica, Universidade de São Paulo, 2009.Flat steel cold rolling is a complex industrial process, carried out into a very harsh environment, characterized by uncertain parameters, large delay times, multiple inputs and outputs, and having a strong interaction between the diverse variables (GUO, 2000). The control system of such a process obtains its setpoints from a mathematical model, responsible for process optimization. Given process peculiarity, the mathematical model must have adaptive features, that is, its parameters need to be continuously adjusted, on the basis of actual operational results. Adaptation is an essential characteristic for achieving product quality requirements and, as a consequence, for the plant feasibility itself. First of all, this work surveys the strategies commonly used in the development of cold rolling adaptive models, pointing out their strengths and deficiencies. Following to this, aiming at the minimization of such deficiencies and by this way improving the performance of the model, this work proposes the application of Computational Intelligence techniques (specifically Artificial Neural Networks) in order to bestow greater effectiveness to the adaptation of two important parameters, namely, the material flow stress and the friction coefficient between rolled strip and rolling cylinders. The text presents the theoretical foundations, the methodology and the implementation of the solution, as so the preliminary results after its application in a real steel mill plant. Finally, the text presents the conclusions, sums up the contributions of the research work and suggests further research to improve the results.