The distributed learning carried out by each organization allows for integration of model parameters without compromising data confidentiality, improving the accuracy of lifetime predictions for heat-resistant materials. Credit: DeMura Masahiko Demura, National Institute of Materials Science
NIM and its collaborators have developed a model designed to predict the long-term durability of a variety of heat-resistant steel materials by performing machine learning while maintaining the confidentiality of each organization's data. This study is published in Tetsu and Hagane.
The privately owned data from private companies is highly confidential and shares organizations with each other for joint research and development. However, interorganizational data collaboration is desirable as it is extremely time-consuming and expensive to generate such data. In particular, it can take more than a decade to obtain lifetime data on heat-resistant materials used in power generation facilities, highlighting the need for industry and public sector cooperation.
NIMS developed a system that enabled multiple organizations (6 private companies and two national R&D institutions) and independently performed machine learning using its own local data, while maintaining confidentiality (i.e. through federal learning).
As a result, we have jointly constructed a “global model” that can predict the long-term durability of heat-resistant steel materials. The global model showed significantly higher prediction accuracy than the local model constructed using only NIMS data. This represents the first example of industry-public sector data collaboration through federal learning.
These results are expected to promote industry and public sector data collaboration across a wide range of materials research fields. The federated learning system developed by NIMS is public and open source. Going forward, NIMS plans to act as coordinators and promote collaboration to meet the growing demand for industry-public sector partnerships.
The federated learning system used in this study was developed and released as open source by NIMS and ELIX.
detail:
Junya sakurai et al., Union learning of creep rupture times and high temperature tensile strength prediction models; Tetsu and Hagane (2025). doi: 10.2355/tetsutohagane.tetsu-2024-124
Provided by the National Institute of Materials Science
Quote: Machine learning models predict the durability of heat-resistant steel while maintaining data confidentiality (June 20, 2025) Retrieved from https://techxplore.com/news/2025-06-machine-resistant-steel-durability-confidentiality.html
This document is subject to copyright. Apart from fair transactions for private research or research purposes, there is no part that is reproduced without written permission. Content is provided with information only.
