About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
Presentation Title |
ML-DiCE: A Machine Learning Framework for Predicting Diffusion Coefficients in Impure Metallic and Multi-Component Alloy Media |
Author(s) |
Arjun S. Kulathuvayal, Yi Rao, Yanqing Su |
On-Site Speaker (Planned) |
Arjun S. Kulathuvayal |
Abstract Scope |
Machine learning (ML) efficiently identifies patterns between complex material properties. This adaptability is particularly useful for estimating, tuning, and designing material properties, outperforming traditional computational methods in complex systems with multiple elements.
Our work introduces ML-DiCE (Machine Learned Diffusion Coefficient Estimator), an ML framework that accurately predicts diffusion coefficients in impure metallic (IM) and conventional multi-component alloy (MCA) media. ML-DiCE integrates five specialized predictive models for distinct diffusion modes: impurity and self-diffusion in IM media, and self, impurity, and chemical diffusion in MCA media. These models utilize chemical descriptors of diffusion medium and diffusing element, along with temperature, and are trained using random forest and neural networks.
With an R2 score above 0.90 and an MSE below 10−16, ML-DiCE demonstrates high predictive accuracy. This research also highlights the potential featurization scheme of alloys in predicting diffusion processes in complex metallic systems, aiding in the design of new materials. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, Mechanical Properties, High-Entropy Alloys |