Research of a Multi-Output Regression Model for Roll Gap Prediction in Roll-to-Roll Calendering of Dry Battery Electrodes

This study develops a multi-output regression model to predict multi-stage roll gaps in a roll-to-roll calendering process for dry lithium-ion battery electrodes. We collected 2,912 samples from process logs measured across Calender 1-6 and used 33 variables (line speed, roll temperature, roll pressure, and web tension) as inputs. Ten OS/DS gaps of Calender 2-6 were predicted simultaneously using a MultiOutputRegressor, and MLR, RandomForest, and XGBoost were compared with 5-fold cross-validation. Tree-based nonlinear models outperformed the linear baseline, and XGBoost achieved the best accuracy with stable generalization. Gap-wise evaluation and feature-importance analysis identified difficult sections and highlighted line speed, roll pressure, and roll temperature as dominant factors. The proposed model can act as a virtual sensor to fill measurement gaps and estimate the full gap profile in real time, supporting monitoring, gap-control strategy design, and operating-condition optimization.