About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Aluminum Reduction Technology
|
| Presentation Title |
Anode Effect Detection in Aluminum Reduction Cells Based on Pruned Spectral Graph Clustering and Multi-Zone Collaboration
|
| Author(s) |
Zhifang Zhang, Jun Tie, Rentao Zhao, Yi Meng, Chun Li |
| On-Site Speaker (Planned) |
Zhifang Zhang |
| Abstract Scope |
The anode effect (AE) significantly affects the efficiency and energy consumption of aluminum reduction, yet voltage-based detection can hardly warn while the voltage is still normal. This study proposes a multi-zone collaborative detection method based on Pruned Spectral-Graph Coordination (PSGC) of zone currents: with a pruning threshold of |r|≥0.8, ten interpretable spectral features are extracted, notably the pruned co-active zone count. On 33 held-out high-voltage events, an absolute voltage-baseline rule detects effects a median ≈42 s before the 8000 mV crossing (≥30 s in 64% of the events it detects). Using the co-active-zone count as a spectral confirmation cuts false alarms from 30 to 11 (5-min merged) (alarm share ≈8%→1%) while keeping early warning on most events, and XGBoost/CatBoost trained on the same early-onset labels reach a comparable early warning. The lightweight computation permits real-time deployment on the cell controller; the evaluation is leakage-free. |
| Proceedings Inclusion? |
Planned: Light Metals |
| Keywords |
Aluminum, Machine Learning, |