Feasibility Study on Wrinkles Detection During Web Transportation Using Convolution Neural Network

Abstract

A roll-to-roll web handling system to transport the web typically comprises an unwind/rewind roller, guide rollers, drive rollers coupled onto servo motors, and a processing part. In the system, web tension and conveying speed controls play an important role in improving product quality. On the other hand, web wrinkles caused by misalignment become a problem because many guide rollers are used in the system. This paper demonstrates a feasibility study of web wrinkle detection based on image processing and deep learning during web transportation in web handling systems. The prototype used in this study can generate arbitrary misalignment between rollers by connecting a drive unit to a guide roller. The camera is used to capture the wrinkles on the web. The web wrinkles are detected using a convolutional neural network based on the captured image. The validity of the proposed approach has been verified by conducting the experiments using the prototype.