About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Melt Processing, Casting and Recycling
|
Presentation Title |
Enhancing Quantification of Inclusions in PoDFA Micrographs Through Integration of Deterministic and Deep Learning Image Analysis Algorithms |
Author(s) |
Anish K. Nayak, Hannes Zedel, Shahid Akhtar, Robert Fritzsch, Ragnhild E. Aune |
On-Site Speaker (Planned) |
Hannes Zedel |
Abstract Scope |
Assessing the cleanliness of aluminum melts has traditionally relied on manual labor, which is both time-consuming and subjective. The present study builds upon previous work in digital image analysis to quantify inclusions in PoDFA micrographs. By combining deterministic methods, unsupervised machine learning, and neural networks, cleanliness data comparable to PoDFA assessments have been achieved. A major challenge with neural networks is generating sufficient training data with accurate labels. To address this, the previously suggested approach has been enhanced by incorporating refined isolation strategies for target classes, resulting in higher-quality training data and improved prediction accuracy of the neural network. Additionally, the post-processing of neural network predictions has also been enhanced. The present findings demonstrate that the integration of different digital image analysis methods yields more reliable cleanliness data compared to previous implementations. The present integrated approach offers a promising alternative to manual PoDFA assessments, improving efficiency and reducing human biases. |
Proceedings Inclusion? |
Planned: Light Metals |
Keywords |
Aluminum, Characterization, Machine Learning |