About this Abstract |
| Meeting |
TMS Specialty Congress 2026
|
| Symposium
|
4th World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2026)
|
| Presentation Title |
Accelerated Prediction of EAF Melting Processes Through an Xgboost-Based Reduced-Order Model |
| Author(s) |
Shiyu Wang, Orlando Ugarte, Sunday Abraham, Yufeng Wang, Randy Petty, Hong Wang, Tyamo Okosun, Chenn Q. Zhou |
| On-Site Speaker (Planned) |
Shiyu Wang |
| Abstract Scope |
High-fidelity computational fluid dynamics (CFD) simulations are used to describe the scrap melting in a DC Electric Arc Furnace (EAF). However, the computational cost of CFD simulations limits their applicability in real-time optimization and control. This work presents a machine-learning–based reduced-order model (ROM) using Extreme Gradient Boosting (XGBoost) trained on a CFD dataset including variations in arc power of a real DC-EAF operation provided by SSAB. The training data captures the time evolution of critical melting parameters, including scrap mass, molten steel mass, and bath temperature, enabling the ROM to learn the nonlinear dynamics of the melting process. In addition, the ROM is extended to predict full heat progression using only initial furnace conditions through a recursive forecasting approach. Results show that the XGBoost-based ROM closely matches CFD predictions for key variables while cutting computation time from hours to milliseconds, allowing for near-real-time evaluation of melting behavior. |
| Proceedings Inclusion? |
Undecided |