About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Advances in Refractory High Entropy Alloys and Ceramics
|
| Presentation Title |
Improving the Use of Machine Learning Tools for High Entropy Alloy Development |
| Author(s) |
David Flores, Wesley Reinhart |
| On-Site Speaker (Planned) |
David Flores |
| Abstract Scope |
The compositional complexity of High Entropy Alloys (HEAs) has prompted the use of machine learning (ML) models to elucidate their composition-processing-property relationships. 'Interpretable’ ML techniques, like SHAP, reveal what inputs these models prioritize, enabling the development of design insights that focus experimental efforts.
This work explores an updated approach for implementing interpretable ML in the specific context of HEA design. The commonly used SHAP tool is sensitive to correlations which are often present in HEA datasets, especially those derived from experiments. We present the use of the corrSHAP tool, an extension of the established SHAP method that accounts for these correlations, thus respecting the physical and experimental constraints of HEA data.
This method yielded more accurate insights into composition-property relationships, highlighting otherwise underestimated elemental contributions that can maximize HEA mechanical properties. This enables better HEA design principles and can guide the exploration of both composition and processing parameters. |