aiMP and aiOQ are databases derived from the 0 K density functional theory (DFT) calculations data stored in the Materials Project and Open Quantum Materials Database (OQMD) repositories, respectively. aiMP and aiOQ databases rely on methods to process 0 K DFT data using machine learning models trained on thousands of compounds. These models adjust formation enthalpies to improve consistency with existing CALPHAD (CALculation of PHAse Diagrams) databases and predict thermodynamic properties such as entropy and heat capacity as functions of temperature.
This work demonstrates three Materials Informatics applications of large-scale CALPHAD-compatible databases enabled by automated workflows.
First, a comparison was made between the SGTE Pure Substance database (SGPS), containing 3927 compounds, and the aiMP database, which includes overlapping entries for 1519 compounds. For these overlapping compounds, the enthalpy of formation, entropy at 298 K, and heat capacity at 298 K were analyzed. Any discrepancies exceeding the inherent error of the machine learning models were flagged. A literature survey was then conducted for compounds with larger discrepancies and erroneous data was confirmed in approximately 0.7% of the SGPS data.
Second, the aiMP database was used to estimate phase diagrams and identify potential new coating materials for SiC/SiC composites, which are under investigation as accident-tolerant fuel cladding materials.
Finally, it is shown that aiMP can serve as a starting point for both traditional and automated CALPHAD modeling. Three examples were explored Al-Ca, Mg-Si, and Ca-Li. These examples highlight the versatility of machine learning-enhanced thermodynamic databases in accelerating material discovery and improving database reliability.
扫码关注我们
求助内容:
应助结果提醒方式:
