About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Aluminum Reduction Technology
|
| Presentation Title |
Linking Refractory Degradation and Failure Modes in Hall–Héroult Cells through Integrated Autopsy and Machine Learning Analysis |
| Author(s) |
Mahesh Sahoo, Jayesh Pawar, Hansraj Singh, chittaranjan Sarangi, Venkannababu Thalagani |
| On-Site Speaker (Planned) |
Mahesh Sahoo |
| Abstract Scope |
Refractory lining performance critically influences the life and efficiency of Hall–Héroult aluminium reduction cells. This study integrates pot autopsy, material characterization, and machine learning (ML) to identify degradation mechanisms. Autopsy revealed distinct cathode zones linked to sodium ingress, aluminosilicate transformation, and NaF-induced expansion. XRD/XRF confirmed nepheline and albite phase formation, while physical measurements showed increased porosity and reduced density, indicating structural weakening.
ML classifiers trained on key operating parameters achieved ~95% accuracy in distinguishing failure modes (Plan stoppage vs. side leakage). Feature analysis identified bath temperature fluctuations and metal height instability as dominant contributors, consistent with autopsy observations. Regression models for life prediction showed limited accuracy, highlighting the need for material-level inputs.
The approach establishes a physics-informed link between operating conditions and failure behavior for improved refractory design and diagnostics. |
| Proceedings Inclusion? |
Planned: Light Metals |
| Keywords |
Aluminum, Characterization, Machine Learning |