About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Computational Thermodynamics and Kinetics
|
| Presentation Title |
Predicting Mixing Enthalpy of Refractory Multicomponent Alloys by Physics-Informed Machine Learning |
| Author(s) |
Shreyash Ghorpade, Christopher Tandoc, Yong-Jie Hu |
| On-Site Speaker (Planned) |
Shreyash Ghorpade |
| Abstract Scope |
Spinodal decomposition offers a unique strengthening mechanism in refractory multi-principal element alloys (RMPEAs), fundamentally governed by the topology of the mixing enthalpy curve therefore, accurately mapping this regime is critical. A key motivation for this work is the significant discrepancies identified between a mainstream CALculation of PHAse Diagrams (CALPHAD) database and Density functional Theory (DFT) calculated mixing enthalpies in several representative binary systems. Since the CALPHAD approach relies on binary and ternary interaction parameters for extrapolation to higher-order systems, such inaccuracies undermine the reliability of CALPHAD-based phase stability assessments for RMPEAs. A physics-informed gradient boosting machine learning framework is proposed which is trained in binary and ternary alloy DFT data. Utilizing physics-based descriptors that encapsulate intrinsic electronic interactions, with the aim to efficiently extrapolate the accurate contours into complex quaternary and quinary composition spaces. This circumvents traditional computational bottlenecks, enabling the accelerated design of spinodally-strengthened RMPEAs.
|
| Proceedings Inclusion? |
Planned: |
| Keywords |
High-Entropy Alloys, Computational Materials Science & Engineering, Machine Learning |