Improved control of electric arc furnace operations by process modelling

The four project partners have modelled the electric arc furnace-based steelmaking process using four different approaches. The employed techniques are: artificial neural networks for energy demand and end-of-heat modelling; neuro-fuzzy and 'ApiMars'-based steel temperature and end-of-heat modelling; artificial intelligence- based heat evolution classification and predictive control; and chemo-physical calculation of scrap melting and heat evolution taking into account a series of furnace parameters. All models aimed at a better understanding of the complex steelmaking process and a better appraisal of the end-point of the heats in order to help the operator to finish the heat at the right moment without wasting power-on time and energy. Three of these models have been implemented online at three different electric arc furnaces: ACB in Spain, Carsid in Belgium, and ProfilArbed Esch-Belval in Luxembourg. The novel predictive control model has been tested offline for the electric power input variable to the furnace. Although largely different approaches were used, the obtained dynamic models describe the process with similar accuracies in terms of end-of-heat prediction and steel tapping temperatures obtained: errors in the range of 5 to 8 kWh/(tonne liquid steel) correspond to final steel temperature predictions within 20 to 25 °C. The online implementations are in daily use at the three steel plants, where they help to increase the productivity of the electric arc furnaces.