Domestic Researchers Propose Platform to Shorten New Material Development from 20 Years to 5 Years

KAIST Professor Hong Seungbeom's Team Develops Algorithm Utilizing Machine Learning and Artificial Intelligence

[Asia Economy Reporter Kim Bong-su] Domestic researchers have proposed a platform that can shorten the development of new materials, which typically takes 20 years, to 5 years by utilizing machine learning, artificial intelligence, and three-dimensional multi-scale printing technologies.

The research team led by Professor Hong Seung-beom of the Department of Materials Science and Engineering at the Korea Advanced Institute of Science and Technology (KAIST) announced on the 31st that they developed an algorithm that dramatically shortens the period from new material design to market entry by combining multi-scale multi-mode imaging technology and machine learning techniques to derive high-dimensional structure-property and process-structure correlations, while also utilizing artificial intelligence and 3D multi-scale printing technology.

This research is part of KAIST’s top 10 flagship projects and the global singularity initiative, the "KAIST New Materials Revolution: M3I3 Initiative." The M3I3 platform presented in the paper by the research team is expected to be applicable not only to high-capacity energy material design but also to high-density memory materials and high-performance automotive and aerospace materials. The study was published online on February 12 in the international academic journal ACS Nano.

The core of this research is that by integrating multi-scale and multi-mode imaging technology, data mining and machine learning, and multi-scale manufacturing technology, it becomes possible to reverse-engineer future-required new materials and rapidly secure process recipes. For example, the reason why Goryeo celadon cannot be reproduced today is that the artisans of the Goryeo era did not leave behind their secrets. However, in the future, if the multi-scale structure of Goryeo celadon is imaged and digitized, and the process to realize the structure is reverse-engineered using machine learning, reproducing Goryeo celadon could become possible.

In this paper, to verify the effectiveness of the M3I3 platform, research was conducted applying it to battery materials. To prove that the development period of high-capacity battery materials can be shortened, over 50 students read 20 years’ worth of academic papers and extracted data to derive correlations between the energy density of cathode materials and material composition. Then, using machine learning techniques, a model was established based on process, measurement, and structural variables found in the papers. The model’s accuracy was measured by synthesizing under random conditions, demonstrating the excellence of data mining and machine learning.

Using various microscopes such as Transmission Electron Microscopy (TEM), Scanning Transmission Electron Microscopy (STEM), Atomic Force Microscopy (AFM), optical microscopy, and spectroscopic equipment including X-ray, Raman, and UV/Visible/IR, the platform derives multi-scale structure↔property correlations based on the obtained images and spectral data, and collects various process variable data to establish process↔structure correlations. This is a key core of the M3I3 platform.

In particular, by integrating experimental data and simulation data, and converting machine learning-generated virtual data into meaningful big data according to scientific standards, it becomes possible to develop a reverse-engineering algorithm connecting property→structure→process using machine learning. Through this, process recipes for new materials with properties needed in the future can be rapidly secured.

Professor Hong Seung-beom said, "With technological advances, we have entered an era where not only the visible shape and structure of materials but also invisible structures can be seen, and even properties can be visualized as functions of space and time. By combining new material imaging technology and machine learning technology and elevating 3D printing technology to multi-scale automatic synthesis technology, the 20-year development period for new materials can be shortened to within 5 years."

This content was produced with the assistance of AI translation services.