Blast Furnace Control by Artificial Intelligence

Prediction and Explanation Models for Hot Metal Temperature, Silicon Concentration, and Cooling Capacity in Ironmaking Blast Furnaces

“…In our application scenario, all analyzed data were recorded at blast furnace 6 of voestalpine Stahl GmbH. It is a nonpressurized furnace with an inner volume of about 1200 m 3 and an average production of some 100 t of hot metal per hour.…”

“…Due to the lack of thermodynamic and kinetic data for very high temperatures and the complex dependencies between internal system variables and the three targets under examination, it is impossible to establish analytical models based on physical and chemical reactions backgrounds only. Expert systems have been established as promising alternative in the past for blast furnace modeling purposes (ranging back to the end of 1980s and beginning of 1990s), see several pioneering works in the studies by Amano et al, Iida et al, and Tianjun et al [2][3][4] Therein, knowledge is coded based on experience by experts working with the process since several years and thus they allow explainable insights into the process and (predicted) states. In the study by Amano et al, [2] the artificial and logical intelligence system (ALIS) for blast furnace operation embeds expert rules for a proper supervision of the process.…”

“…The expert rules provide an interim hypothesis for each item under supervision, such as the gas flow distribution or the furnace heat level, whereas a final decision of proper action is made based on the results of the interim hypothesis judgment. In the study by Iida et al, [3] expert rules are coded in a rule book to perform an analysis of the current process situation, a recognition of phenomena and the determination of proper actions. Therefore, several process variables such as hot metal temperature or heat consumption are used, but no future predictions of these variables are made for early reaction purposes.…”