Research News

AI alloy prediction process

By combining experimental and computational data with AI, new metals could be discovered.

Credit: Osaka Metropolitan University

Metals can be transformed into alloys by adding different elements, allowing their strength, stability, and other properties to be tailored. A key parameter for understanding these properties is the volume size factor (VSF), which quantifies lattice distortion caused by differences in atomic size.

However, measuring VSF experimentally is both time-consuming and costly, leaving many alloy systems without reliable data. Although VSF can also be predicted using first-principles calculations, significant discrepancies between calculated and experimental values have been reported for certain alloy systems.

A research team led by Professor Tokuteru Uesugi at Osaka Metropolitan University’s Graduate School of Informatics constructed a large-scale VSF database based on first-principles calculations for 1,998 binary solid-solution systems and developed an AI-driven transfer-learning method that corrects discrepancies between calculated and experimental VSF values. The approach improved prediction accuracy for alloys with available experimental data while demonstrating reliable predictive performance for previously unexplored alloy systems.

"We expect that this achievement will make it possible to identify promising alloy candidates on a computer before conducting costly experiments, even for alloy systems that have been difficult to investigate experimentally," said Professor Uesugi. "This approach could help shift materials development away from the traditional trial-and-error process of 'make and test' toward a more efficient paradigm of 'predict before making.' As a result, it has the potential to reduce both development time and experimental costs."

Paper Information

Journal: Materialia
Title: Transfer-learning-based refinement of volume size factors from first-principles calculations
DOI: 10.1016/j.mtla.2026.102888
Author: Tokuteru Uesugi and Shunto Matsuura
Published: 4 September 2026
URL: https://doi.org/10.1016/j.mtla.2026.102888

Contact

Tokuteru Uesugi

Graduate School of Informatics
Email: uesugi[at]omu.ac.jp

*Please change [at] to @.