Expanding materials selection via transfer learning for high-temperature oxide selection

TL;DR

This paper develops transfer learning models to predict melting temperatures of oxides, enabling the identification of high-temperature materials for advanced applications using limited data and surrogate computational methods.

Contribution

It introduces a multi-step sequential learning approach leveraging surrogate data to predict oxide properties with small datasets, expanding materials selection capabilities.

Findings

Predicted melting temperatures for nearly 11,000 oxides.

Quantified uncertainties in property predictions.

Demonstrated effectiveness of transfer learning in materials discovery.

Abstract

Materials with higher operating temperatures than today's state of the art can improve system performance in several applications and enable new technologies. Under most scenarios, a protective oxide scale with high melting temperatures and thermodynamic stability as well as low ionic diffusivity is required. Thus, the design of high-temperature systems would benefit from knowledge of these properties and related ones for all known oxides. While some properties of interest are known for many oxides (e.g. elastic constants exist for over 1,000 oxides), melting temperature is known for a relatively small subset. The determination of melting temperatures is time consuming and costly, both experimentally and computationally, thus we use data science tools to develop predictive models from the existing data. The relatively small number of available melting temperature values precludes the use of standard tools; therefore, we use a multi-step approach based on sequential learning where surrogate data from first-principles calculations is leveraged to develop models using small datasets. We use these models to predict the desired properties for nearly 11,000 oxides and quantify uncertainties in the space.