About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Interpretable Data-Driven Modeling of Composition-Dependent Tensile Response of Multicomponent Alloys |
| Author(s) |
Bekassyl Battalgazy, Daniel Salas Mula, Ibrahim Karaman, Raymundo Arroyave, Ankit Srivastava |
| On-Site Speaker (Planned) |
Ankit Srivastava |
| Abstract Scope |
The mechanical behavior of single-phase multicomponent alloys is strongly influenced by their chemical composition. Traditional constitutive modeling approaches in general do not include compositional effects and thus require a distinct constitutive model for each alloy, limiting generalizability and interpretability. In this work, we present a data-driven framework for discovering interpretable, composition-dependent models to represent tensile response using symbolic regression. The framework is applied to a dataset of 76 multicomponent alloys, and composition-dependent constitutive models are identified and their performance evaluated. The results show that a composition-dependent constitutive model with as few as 14 parameters can accurately capture the tensile response of these alloys. We also carried out a rigorous analysis of model complexity and predictability as well as implications of underfitting and overfitting in the dataset. Furthermore, the dataset is also analyzed using a comparable neural network model, and the results are compared. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Mechanical Properties, High-Entropy Alloys |