| Title |
AI-based Quality Prediction and Inverse Design Optimization for NCM811 Cathode Material using Raw Material CoA |
| Authors |
김수림(Soo-Rim Kim) ; 송정윤(Jung-Yoon Song) ; 이용귀(Yong-Kwi Lee) ; 오현우(Hyun-Woo Oh) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.6.141 |
| Keywords |
Cathode material; NCM811; CoA; Inverse design; Quality prediction |
| Abstract |
This study developed a machine learning prediction model based on raw material Certificate of Analysis (CoA) data and applied an inverse design methodology to address lot-to-lot quality variation in lithium-ion battery cathode material manufacturing. A total of 802 datasets were split into training and test sets at an 8:2 ratio, and 13 key input variables were selected through Pearson correlation analysis. Comparing four algorithms including H2O AutoML, XGBoost, and Random Forest, H2O AutoML achieved the highest predictive accuracy with an average test R² of 0.865. Calibration functions modeling the relationship between predicted and actual values were derived for each target variable with GradientBoosting showing the best performance across y005, y006, and y008, the calibration R² for y006 reached 0.960. Using the trained prediction model, a four step inverse design algorithm incorporating the Winsorization concept was applied to derive 60 optimal raw material combinations that minimize quality dispersion. The inverse design results showed substantial dispersion reduction on a predicted value basis with standard deviations decreasing by 98.5% for y005, 83.3% for y006, and 91.9% for y008, and on an actual value basis, y005 and y008 showed reductions of 14.0% and 45.5% respectively. This study presents a data driven inverse design methodology capable of ensuring quality uniformity at the raw material input stage without any modification to process conditions. |