AI & Materials
Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.
매일 00·06·12·18시(한국시간)에 웹에서 새 소식을 찾습니다. AI 활용 방법과 결과가 확인된 새 사례만 게시합니다.
Checking notifications…
A study dated 2026-09-14 reported an approach that used interpretable machine learning to rapidly narrow down metal-organic framework (MOF) candidates that could exhibit altermagnetism. It was used for predicting the magnetic properties of MOFs and screening candidate materials, but the extent of experimental validation and generalizability cannot be judged from the title and list information alone.
A study reported the development of a large language model-based pipeline that automatically extracts, structures, and builds a database from composition, process, and property attributes in concrete and blended cement paper text. Results included a maximum F1 of 0.98, about 9,000 records within 1 hour, and extraction of more than 100 attributes.
An AI-assisted workflow has been reported that links segmentation, FFT peak detection, crystal phase identification, strain map calculation, and 3D model generation from electron microscopy (S/TEM) images. By claiming to reduce analysis that previously took experts days to minutes, it has direct significance for automating the processing and interpretation of materials microscopy data. However, reproducibility across individual samples, instruments, and image quality still requires separate validation.
A 2026-09-11 paper in npj Computational Materials used multi-agent LLMs and in-context learning to operate an X-ray nanoprobe beamline and an autonomous robotic station for materials design. The key point was evaluating the possibility of on-the-job learning that orchestrates complex measurement workflows and incorporates expert feedback.
A 2026-09-11 paper in Nature Communications built an autonomous laboratory for data-driven homogeneous catalysis and used Bayesian optimization to closed-loop optimize ligands and reaction conditions. More than 360 experiments improved search efficiency, and the authors also reported condition-dependent selectivity inversion and scale-up validation.
The study published in ACS Central Science on 2026-09-04 used a multi-agent large language model framework to explore and design oxygen reduction reaction (ORR) candidates for single-atom catalysts. Multiple LLMs with divided roles collaborated to identify high-performance candidates, and the authors introduced this system as MAESTRO.
It was reported that by using active learning to explore the compositions of multicomponent Invar alloys and screen candidates, a material with low thermal expansion, high Curie temperature, and excellent phase stability was discovered. This can be interpreted as a case in which active learning was used to reduce the candidate space in materials discovery.
Using a machine learning potential, a global structure search was performed for 11 Bi-based binary systems in the 0–20 GPa range, and several unreported compounds were found. Previously experimentally known compounds were also reproduced in the search space, supporting the robustness and reliability of MLP-based structure search.
A machine learning classifier reflecting crystal-field information rapidly screened more than 10,000 candidates and proposed 15 two-dimensional magnets with perpendicular magnetic anisotropy (PMA). The reported classification accuracy was 89%, and Ba₂VTe₂S₂ was proposed as a promising candidate.
A study in npj Computational Materials presented a method that uses Bayesian optimization to automatically reconstruct surface structures from low-energy electron diffraction (LEED) data. It was introduced as an automated approach to accelerate surface structure interpretation, while detailed quantitative results are limited to what is available in the public text.
ORNL officially introduced research in which AI autonomously carried out atom-scale material assembly and synthesis. The announcement is based on a study published in ACS Nano and shows an approach for automating customized materials fabrication.
A split MLIP that separates intralayer and interlayer interactions in layered van der Waals materials, together with a data curation approach, was presented. In large-scale moiré-structure prediction and verification, energy and force prediction improved by about 10 times over existing models, and a physical verification metric using the distribution of stacking configuration was also proposed.
EPFL researchers used a machine learning interatomic potential (MLIP) to reproduce the early stage of solid-state precipitation and phase-transition competition in dilute Mg–Nd alloys through large-scale atomistic simulation. They simulated the complex early precipitation behavior observed in experiments and suggested competition between order-disorder and structural transformation, as well as the possibility of a continuous hcp↔bcc transition.
AutoEMX, introduced in Nature Communications, combined an automated SEM-EDS workflow with machine learning to correct particle-shape artifacts in multiphase powders and rapidly quantify individual compositions. The study reported reducing composition quantification errors to the 5~10% level.
Nature Communications introduced a machine learning-based virtual reference electrode that estimates the anode potential from standard cell signals alone and detects lithium-plating risk, extending cycle life by 8.25 times. This is an example of direct AI application to battery management and safety diagnostics.