Deep Learning for Estimating the Fill-Level of Industrial Waste Containers of Metal Scrap: A Case Study of a Copper Tube Plant

• Published in: Applied sciences Vol. 13; no. 4; p. 2575
• Main Authors: Alexopoulos, Kosmas, Catti, Paolo, Kanellopoulos, Giannis, Nikolakis, Nikolaos, Blatsiotis, Athanasios, Christodoulopoulos, Konstantinos, Kaimenopoulos, Apostolos, Ziata, Efstathia
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.02.2023
• Subjects: Automation; Case studies; Classification; Computer vision; Containers; convolutional neural networks; Copper; Copper products; Data collection; Deep learning; digitalization of industry; Estimation; Industrial applications; Industrial wastes; Industry 4.0; Inspection; Internet of Things; Machine vision; Management systems; Mechanical systems; Metal scrap; Model accuracy; Neural networks; Process management; Refuse and refuse disposal; scrap management; Scrap metals; Sensors; Vision systems; Waste containers; Waste management
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app13042575

Advanced digital solutions are increasingly introduced into manufacturing systems to make them more intelligent. Intelligent Waste Management Systems in industries allow for data collection and analysis to make better-informed decisions, monitor and manage processes remotely, and improve waste management. In many industries, scrap is collected in large waste containers located on the factory floor, usually close to its source. In most cases, monitoring of waste containers’ fill levels is either manually performed by visual inspection by the operators working in close proximity or by employing intrusive mechanical systems such as weight sensors. This work presents a computer vision system that uses Deep Learning (DL) and Convolutional Neural Network (CNN) for the automated estimation of the fill level in industrial waste containers of metal scrap. The training method and parameters as well as the classification performance of VGG16 CNN that was retrained upon images collected in the field, are presented in detail. The proposed method has been validated upon an industrial case study from the copper tube production industry in which the fill level of two waste containers is estimated. A total of 9772 images were captured for the first container and 11,234 images for the second container. The VGG16 model achieved an accuracy from 77.5% to 95% on the testing dataset. The industrial case study demonstrates that the proposed computer vision system has sufficient accuracy for classifying the fill levels of metal scrap containers which allows for the development of waste management applications in industrial environments.