Machine Learning-Based Splice Detection Utilizing Web Edge Sensor Data

Kevin Lifsey

Aravind Seshadri, Roll-2-Roll Technologies

Splices constitute a common disturbance in all Roll-to-Roll (R2R) machines, produced during the joining of two web rolls. Vendor splices, produced in upstream processes, are traditionally tracked using dedicated splice detection sensors, which monitor the splice location within the converting machine during web transportation. Due to variations in splice joint types, transporting a splice within an R2R machine often incites both tension and lateral disturbances. Tracking the splice location is therefore essential to mitigate these effects. However, splices ultimately mar the aesthetic of the final product, leading to their removal via automated defect rejection systems.

Splices sometimes employ specialized tapes, varying in color or metallic properties, to ensure detection by splice sensors. Despite this, the array of splice tape types frequently results in missed vendor splices by suppliers, even with specialized sensors.

This paper explores the potential for utilizing edge position data in splice detection. Splices represent a discontinuity between two webs, a characteristic potentially observable through the continuity of edge position information. However, the absence of this discontinuity in perfectly made splices may be difficult to detect through edge position data. Therefore, this paper applies machine learning methodologies to demonstrate splice detection using edge position sensor data.

Through a series of experiments with varying tape types (color and quality), the efficacy of the proposed machine learning algorithm for real-time splice detection is assessed. These experiments, conducted at different web speeds across a wide array of materials, validate the utility of our proposed approach in a diverse range of scenarios.

This post is for paying members only


Already have an account? Log in