A Machine Vision System using Machine Learning for Tension Monitoring of Metal Sheets in Small-Scale Roll-to-Roll Process

This paper proposes a machine vision system to monitor the tension in the metal sheet on a small-scale roll-to-roll machine. The proposed tension monitoring is applied to a roll-to-roll chemical vapor deposition (R2R-CVD) process for the continuous production of graphene, in which the copper sheet is fed through a high-temperatured vacuum chamber. However, traditional contact-type metal tension sensors have limitations to the harsh environment inside the sealed chamber. Therefore, this study proposes a non-contact machine vision system to measure metal tension by analyzing the curvature of the copper sheet rolling. A key challenge lies in the highly reflective surfaces of both the metal sheet and the chamber. To address this, this paper designed a specific light source and image processing algorithm for accurate segmentation of the reflective metal sheet. In addition, perspective warp and catenary model fitting are employed to estimate the metal curvature. Finally, a machine learning model was trained to classify the tension level from the estimated curvature parameters. The experimental results demonstrate that the proposed system can monitor the curvature of the metal sheet with a high precision of 99.8%.