Modelo de control del ancho de las bobinas laminadas en caliente en el tren reversible mediante técnicas Data Mining

Resum

The control of the dimensional factors in steel making is becoming an increasingly complex task because of the strict requirements for high productivity, low energy consumption, and production of different high quality steel grades. In this paper, data based methods as multivariate adaptive techniques and neural networks have been used to learn and train process parameters and their degrees of influence over width deviations. The artificial intelligence techniques have also been used to check and validate the collected data. A total of 200 variables were considered for the analysis, selecting the relevant variables by means of a semiautomatic strategy gbased in the combination of Self-Organized neural networks and multivariate adaptive techniques. The variables selected, relative to the thermomecanic properties of the slabs and the rolling forces, were used to create a width model to improve the width control in the roughing mill. The model is created as a combination of data based and analytical process models, where the intelligent component plays the role to fit the general physical models to the needs of the automation of a specific mill. The system was tested in the installations of Aceralia (Spain) with a save in material losses and an acceptance rate in customer requirements of a 99.5%.