|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Hume-Rothery Award Symposium: Computational Thermodynamics and Its Implications to Kinetics, Properties, and Materials Design
||Semi-automated CALPHAD Modeling of Alloy Systems
||Richard Otis, Brandon Bocklund, Zi-Kui Liu
|On-Site Speaker (Planned)
The CALPHAD tradition has relied on the experience of experts to pass their wisdom to the next generation of modelers in the form of ad-hoc heuristics and supervised trial-and-error. This approach has been very successful but suffers from poor scalability, and little work has been done in developing tools for extracting insights from thermodynamic data in a scalable, semi-automated fashion. If a CALPHAD model cannot be reproduced, then it cannot be easily improved, making extrapolations and modifications to multi-component databases more difficult.
This work describes two open source software packages, pycalphad and ESPEI, for automating the CALPHAD modeling process. Using machine learning techniques, these packages require little input from the user, making it possible to be integrated into high-throughput modeling infrastructure. With a more robust method for updating multi-component thermodynamic databases based on the latest available data, thermodynamic studies of multi-component materials systems will become more predictive and accurate.
||Planned: Supplemental Proceedings volume