Otimização do processo de fusão e refino primário através de análises estatísticas de parâmetros elétricos e de processos em forno elétrico a arco

Abstract

In semi-integrated steel plants that have electric steelmaking processes, called mini-mills, the production of crude steel occurs through the process of melting a metallic solid load composed of scrap from different origins and/or solid pig iron as a general rule (in some plants, it is possible to find the addition of molten pig iron) - in the metallurgical reactor known as Electric Arc Furnace (EAF), where the primary refining of the melted load also occurs. The process, which originated in the XIX century, has significantly increased its participation on world's steel production in the last decades due to its flexibility of raw materials and product mix, environmental benefits from the recycling of scrap and logistic gains due to its compaction and necessary investments of smaller proportions if compared to the integrated steel mills. Even with the aforementioned advantages, it is necessary to continuously adjust and improve the process in order to operate at competitive costs in the current market, which combines production overcapacity with a slightly growth in demand. Knowing that the EAF process requires a large amount of electricity, optimizations of mass and energy balances are done with high frequency. In general, these calculations are not simple and in some companies there are softwares developed specifically for this purpose. On the other hand, technological evolution has allowed process engineers to capture and store many variables in databases. Aiming to obtain the greatest number of possible responses from the data, it was applied a statistical approach using the multiple linear regression (MLR) and partial least squares (PLS) methods correlating scrap mix, process and electrical parameters of the EAF with the required amount of electrical energy to melt the solid load - one of the largest installments in the production cost of electric steelmakings. The models were evaluated with the actual data of different periods and, in addition, with data obtained by the company's official software for mass and energy balance, showing a good fit with mean errors of less than 5% for all regressions. The statistical model demonstrated a good accuracy for the industrial practice of energy optimization and the advantage of obtaining results from linear equations, besides including factors such as the harmonic distortions that are important measures of load behavior in EAF.