Domestic researchers have successfully applied artificial intelligence (AI) to the lithium-ion battery cathode material production process, reducing the defect rate and increasing the yield.
A research team led by Professor Jeong Im-du of the Ulsan National Institute of Science and Technology designed process conditions that can lower the defect rate of NCM precursors through a joint study with the Organic Materials Research Team at the Pohang Institute of Technology (RIST) and announced this on the 27th.
NCM precursor is a powdery substance mixed with nickel (Ni), cobalt (Co), and manganese (Mn), which is aggregated at high temperatures to create cathode materials for electric vehicle batteries. The higher the nickel content in the precursor particles, the larger the battery capacity, but there is a risk of 'leaching' where nickel does not precipitate properly and remains in the solution or washes away. Leaching leads to defects by causing irregular shapes and composition ratios of the particles, which reduces battery lifespan and performance.
The research team optimized process conditions to suppress nickel leaching and also developed an AI-based real-time facility anomaly detection technology. By controlling the stirring speed of the raw solution containing metal ions, acidity (pH), ammonia concentration, etc., nickel was designed to be placed inside the particles while cobalt and manganese were placed on the outside. When nickel is seated inside the particles, the possibility of leaching decreases, and structural stability increases.
Additionally, the performance of defect detection was significantly improved with domain-adaptive AI technology. Existing AI was only optimized for the conditions learned in the laboratory, meaning that even slight changes in conditions due to facility aging or long-term mass production significantly lowered performance. In contrast, domain-adaptive AI can recognize changes in production environments in real-time and self-correct, enabling stable quality predictions in various situations.
The research team demonstrated this technology in an industrial 11.5-ton reactor. As a result, the number of defective batches was reduced to one-fifteenth of the previous level, and the accuracy of AI-based anomaly detection reached 97.8%. The research team noted that it could reduce raw material and production losses by approximately 2.2 billion won annually.
Professor Jeong Im-du said, "Unlike the small-scale experimental environment in the laboratory, mass production at the site requires significant expense and effort to manage quality and yield, and this AI technology was applied in a real-world setting to induce stable high-quality production," and added, "This can be applied not only to secondary batteries but also across large-scale manufacturing industries such as chemicals, machinery, and semiconductors."
References
InfoMat (2025), DOI : https://doi.org/10.1002/inf2.70031