Post by qoncept

🌬️ During oxygen blowing, everything comes down to one question: Have we reached the target yet? If the answer is yes, oxygen blowing should stop. To answer the endpoint question in real time, we need to know the current state of the melt while the heat is still running, especially the current carbon content. But the bath state cannot be measured continuously in normal EAF operation. Samples are only available at certain points. The process between those points must be estimated. 💡 This is where physics and AI become the key. A physical model based on activity-corrected Gibbs energies captures the metallurgical mechanism: which elements compete for oxygen, and why this competition shifts during the heat. But physics alone carries systematic error. The hybrid approach combines both: xᵢ(hybrid) = xᵢ(ΔG) + εᵢ(ML) In the current implementation, the ML correction is applied to the oxygen share going to Fe. The oxygen shares for the other elements, including C, Si, Mn and Cr, are calculated by the fundamental physical model. That matters because Fe is where the endpoint cost appears most directly. The physical model provides the mechanism. The ML term corrects the residual between predicted and observed Fe oxidation. Developed and evaluated using industrial EAF data from three facilities and ~ 40 000 heats, the hybrid model provides a continuous estimate of bath composition between samples, including the carbon content that matters most. That makes the key question answerable in real time: Have we reached the target yet? And if yes: stop blowing. ➡️ Comment and follow qoncept technology GmbH to stay up to date on how we help transform steelmaking into a more efficient and sustainable industry. #EAF #Steelmaking #Optimization #ArtificialIntelligence

Video Content