About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Verification, Calibration, and Validation Approaches in Modeling the Mechanical Performance of Metallic Materials
|
| Presentation Title |
Understanding Machine Learning Model Predictions for In Situ Material Property Estimation in Digital Qualification Frameworks |
| Author(s) |
Zackary K. Snow, Luke Scime, William Halsey, Chase Joslin, Andres Marquez Rossy, Yousub Lee, Peeyush Nandwana, Wen Dong, Vincent Paquit |
| On-Site Speaker (Planned) |
Zackary K. Snow |
| Abstract Scope |
Recent advances in additive manufacturing process monitoring and analysis techniques have demonstrated an ability to make localized predictions of flaw population statistics, traditionally derived from post-build X-ray computed tomography, using machine learning (ML) models trained on only in situ sensor data and design intent information. These models, called voxelized property prediction models (VPPMs), require a multi-staged learning approach to ensure generalizability across materials, machines, and geometries and must be properly calibrated to be effectively and responsibly leveraged in digital qualification environments. This presentation will review results of an ablation study wherein inputs to the VPPMs are systematically removed to assess their impact on model performance, allowing contributions from each input channel to be quantitatively assessed. Techniques for understanding ML models – generally considered to be “black box models” – will be presented within the context of digital qualification and certification frameworks. This research was sponsored by the DOE EERE AMMTO. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, ICME, Machine Learning |