Data-driven fault detection and positioning of eccentric rolls in roll-to-roll systems using wrap angle and sensor proximity

• FCM method improves fault detection and positioning of eccentric rolls in R2R systems • Optimal feature sets leverage tension data, wrap angles, and sensor proximities • FCM-based features outperform traditional models in eccentric roll detection • 97.3% accuracy achieved while reducing data capacity and processing time • Selective feature optimization enhances efficiency for eccentric roll positioning This paper presents a novel approach for detecting and positioning eccentricity defects in roll-to-roll (R2R) slot-die coating systems, which is critical for maintaining high production quality and efficiency. It introduces the feature combination matrix (FCM) method, which improves fault detection and precise positioning of eccentric rolls through focused feature engineering. The study employs the FCM method to enhance defect detection accuracy, using Support Vector Machine (SVM) as the classifier to consistently evaluate the effectiveness of selected feature sets in identifying and positioning eccentricity in R2R systems. By leveraging tension data, optimal feature sets are identified, emphasizing the Mean, Fast Fourier Transform, and other key variables that capture crucial system dynamics, including wrap angles and sensor proximities. Classification algorithms were used to compare the performance of models employing these optimized FCM-based features against traditional models utilizing a full suite of 40 statistical features. The FCM methodology achieved a notable improvement in eccentric roll detection and positioning accuracy, reaching 97.3%, while also reducing data capacity requirements and processing time. These outcomes highlight the effectiveness of selective feature optimization and strategic combinations, demonstrating that high-impact features enhance both detection accuracy and efficiency. Conclusively, the FCM is positioned as a pivotal tool for advancing industrial diagnostic processes, with future work suggested in the areas of adaptive feature extraction techniques and real-time model integration to improve the reliability and adaptability of R2R manufacturing.