ISSN 2594-357X
34º Seminário de Redução de Minério de Ferro e Matérias-Primas — Vol. 01 , num. 34 (2004)
Title
Authorship
DOI
Downloads
Abstract
In this work we present a system based on techniques of Artificial Intelligence (AI) combined in an unpublished way in literature to predict the silicon grade of samples of pig iron from blast furnace 2 runners in Acesita. Silicon grade is one of the main indicators of pig iron quality, and therefore a system capable of making accurate predictions allows the furnace manager to take corrective action more quickly, thus improving the process quality. The system uses a combination of machine learning, neural networks and bayesian optimization algorithm, and allows online prediction of the silicon grade in the next run of pig iron, using furnace and process information, in addition to the results of mineralogic and chemical analyses of raw materials. The models are exclusively constructed on the basis of historical information from the supervisory and laboratories databases. The system operates with error ranges lower than 5% in normal periods and lower than 15% in periods with anomalies. Tools for relevant data selection are used, without user intervention, which eliminate noise and redundancy of the supervisory data. Such tools explore varied characteristics of the data, and when used together, they generate a consistent data set for use in the AI models. The models are constantly updated through online training evaluations, allowing the automatic incorporation of alterations in the characteristics of the furnace, the process, or the raw materials
In this work we present a system based on techniques of Artificial Intelligence (AI) combined in an unpublished way in literature to predict the silicon grade of samples of pig iron from blast furnace 2 runners in Acesita. Silicon grade is one of the main indicators of pig iron quality, and therefore a system capable of making accurate predictions allows the furnace manager to take corrective action more quickly, thus improving the process quality. The system uses a combination of machine learning, neural networks and bayesian optimization algorithm, and allows online prediction of the silicon grade in the next run of pig iron, using furnace and process information, in addition to the results of mineralogic and chemical analyses of raw materials. The models are exclusively constructed on the basis of historical information from the supervisory and laboratories databases. The system operates with error ranges lower than 5% in normal periods and lower than 15% in periods with anomalies. Tools for relevant data selection are used, without user intervention, which eliminate noise and redundancy of the supervisory data. Such tools explore varied characteristics of the data, and when used together, they generate a consistent data set for use in the AI models. The models are constantly updated through online training evaluations, allowing the automatic incorporation of alterations in the characteristics of the furnace, the process, or the raw materials
Keywords
artificial intelligence, prediction, pig iron, silicon grade, process quality
artificial intelligence, prediction, pig iron, silicon grade, process quality
How to cite
Carvalho, Bernardo Penna Resende de; Senna, André Luiz de; Morais, Fábio Mayrink.
PREDICTION OF SILICON GRADE IN THE PIG IRON USING ARTIFICIAL INTELLIGENCE TECHNIQUES,
p. 447-456.
In: 34º Seminário de Redução de Minério de Ferro e Matérias-Primas,
None,
2004.
ISSN: 2594-357X, DOI 10.5151/2594-357x-34Red-447-456
