Roll misalignment in roll-to-roll (R2R) processes is a critical cause of lateral displacement and tension imbalance, leading to dimensional instability and surface defects in polymer films. Conventional analyses based on beam or camber models often require complex calibration and additional sensing, which can limit their applicability in real-time production environments. This study introduces a diagnostic approach that estimates web misalignment directly from side-to-side tension differences measured in roll-to-roll (R2R) systems. The method eliminates the need for additional sensors and complex geometric calibration, simplifying system setup. The correlation between tension imbalance, lateral displacement, and the equivalent misalignment angle was experimentally established. Our approach produced accurate predictions across various process conditions, including different roll misalignments and applied tensions, and we found that reducing tension “hunting” further enhances prediction stability. This study demonstrates that the proposed tension-based approach can complement existing systems and reduce the reliance on complex external sensing for diagnostic checks of misalignment. By simplifying alignment diagnostics, the method provides a practical route to enhance process setup, reduce downtime, and improve the uniformity of polymer films in continuous manufacturing.

In roll-to-roll slot-die coating systems, backup roll eccentricity induces periodic fluctuations in key process parameters, leading to coating defects. Therefore, accurate prediction of backup roll eccentricity during operation is essential for maintaining stable coating quality. However, conventional methods based on web tension signals have limitations in clearly distinguishing and quantitatively evaluating subtle eccentricities. To address this issue, this study proposes a data-driven framework for predicting backup roll eccentricity using meniscus image information. Representative morphological features are defined to describe the global shape, local shape, and curvature characteristics of the meniscus. Since these features are sensitive to eccentricity-induced process variations, they can serve as effective indicators for eccentricity prediction. The defined features are used to train regression models, and the model with the highest predictive accuracy is selected. Experimental results confirm that the proposed meniscus-based approach significantly outperforms conventional tension-based methods. The proposed method achieves an average Root Mean Square Error (RMSE) of approximately 0.63 μm, an average Normalized Root Mean Square Error (NRMSE) of approximately 0.34, and a coefficient of determination (R²) greater than 0.93 across all test cases. These results demonstrate the feasibility of robust process monitoring using a simple vision sensor configuration.