About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Aluminum Reduction Technology
|
| Presentation Title |
Data-Driven Optimization of AlF₃ Feed Strategy for Enhanced Efficiency in Aluminum Smelter |
| Author(s) |
Saurabh Ghanshyam Kawale, Amit Agashe, Sujit Anandrao Jagnade, Rajshri Magdewar, Parthprasoon Sinha, Praveen Kapse |
| On-Site Speaker (Planned) |
Rajshri Magdewar |
| Abstract Scope |
Maintaining optimal bath temperature is vital in aluminium smelting to ensure high current efficiency and reduce power consumption. Aluminium fluoride (AlF₃) is added for temperature control, but traditional feed strategies often cause deviations, affecting energy use and product quality. This study presents a data-driven approach to optimize AlF₃ feeding using machine learning.
Historical data from over 100 smelting pots across four years were analyzed to develop predictive models for AlF₃ feed rates. Key variables, pot age, alumina feed, voltage, bath temperature, and excess AlF₃, were selected using mutual information techniques. Among several models tested, a random forest model performed best, achieving an R² of 0.93 and MAE of 6.6.
The resulting recommender system provides precise AlF₃ dosing guidance, improving bath temperature stability and reducing feed variation. A 1% gain in current efficiency saves 145 kWh/MT, yielding cost savings of \$4.78/ton, demonstrating the solution’s scalability and industrial value. |
| Proceedings Inclusion? |
Planned: Light Metals |
| Keywords |
Aluminum, Machine Learning, Other |