A joint research team led by Professors Hong Seung-beom and Cho Eun-ae of the KAIST Department of Materials Science and Engineering said on the 26th that it developed a Machine Learning framework that accurately predicts the particle size of battery cathode materials even when some experimental data are missing and presents the confidence level of the predictions.
The most widely used cathode materials in today's electric-vehicle batteries are NCM-series metal oxides that mix nickel (Ni), cobalt (Co), and manganese (Mn). Because this material is directly linked to core performance such as battery lifespan, charging speed, driving range, and safety, precise control of the manufacturing process is important.
The team focused on the fact that the size of the fine primary particles that make up the cathode material is a key factor that determines performance. To that end, they designed an artificial intelligence (AI)-based technology that can more accurately predict primary particle size and efficiently identify process conditions.
Until now, researchers had to perform repeated experiments while changing various conditions such as sintering temperature and time, and material composition to determine particle size. However, in real research settings, it is difficult to measure every combination without omission, and missing experimental values are frequent, limiting precise analysis of the relationship between process conditions and particle size.
The proposed framework combines a technique that supplements missing experimental data by reflecting chemical properties with a probabilistic Machine Learning model that simultaneously computes prediction uncertainty. In other words, it not only outputs a number for what the particle size is likely to be but also provides information on how much that prediction can be trusted.
The team said that after augmenting the experimental data and training, the AI model showed a high prediction accuracy of about 86.6%.
Professor Hong Seung-beom said, "It is significant in that it opens a way to first identify conditions with a high chance of success even without running every experiment," adding, "It will accelerate battery material development and greatly reduce unnecessary experiments and expense."
The research was published in the international journal Advanced Science in Oct. last year.
References
Advanced Science (2025), DOI: https://doi.org/10.1002/advs.202515694