Extraction of Critical Raw Materials From Waste Printed Circuit Boards Using Machine Learning and Computer Vision

The extraction of critical raw materials (CRMs) from waste printed circuit boards (WPCBs) represents a crucial challenge for the economy, particularly in the electronics sector. To improve the concentration of individual CRMs and facilitate their extraction from WPCBs, we propose an innovative approach based on machine learning and computer vision to guide mechatronics system in selectively disassembling various types of electronic components from WPCBs. Our approach addresses the need for a robust and automated solution to identify and localize valuable components that contain CRMs. Specifically, we use YOLOv5 and YOLOv8 models, which are the state of the art convolutional neural networks (CNN) for detection and localization tasks. Our pipeline, called EC-Det, starts from collecting a custom image set called V-PCB, and obtains a large collection of 10,829 high-quality images of WPCBs. These images are used to train the YOLO family variants which show excellent accuracy. Furthermore, we consider not only accuracy but also memory usage, training time, and inference time for our CNN models, evaluating their suitability for resource-constrained on the edge systems. Our results show that integration of machine vision system provides a highly effective and reproducible solution for guiding automated mechatronics system, thereby enhancing the efficiency of the CRMs extraction process.