Content area
Full Text
Abstract- In this study, a deep learning technique has been introduced to enable a novel robotic application in shredded electronic waste (e-waste) sorting. The main objective is the classification of three material groups: circuit boards, plastics, and wires; commonly found in shredded e-waste. Speed and accuracy are the key factors in this process. The desired industry requirement for the segregation purity is 95% in this application. Due to this requirement, we applied a relatively deep model using a combination of the Faster R-CNN algorithm with a ResNet101 feature extractor. Using this combination, we are able to reach an overall purity rate of 98%. Using only one Graphics Processing Unit (GPU) our neural network implementation can infer images at a rate of approximately 20 frames per second. This meets the original requirements for e-waste sorting process in which a conveyor belt carries the shredded pieces at a speed of 1 meter per second. A high-speed parallel robot is utilized to sort the materials into separate bins. The promising result of this study will pave the road to addressing the shortcomings of current e-waste sorting technologies in terms of efficiency and liability.
Keywords: Electronic Waste Recycling; Robotic Sorting; Neural Networks; Machine Learning; Deep Learning.
Submission Type: Regular Research Paper
I. Introduction
In recent decades, consumer electronics have become an integral part of daily life, revolutionizing the way we communicate, retrieve information, and entertain ourselves. Waste management and resource recovery for electrical and electronic equipment, such as computing/display devices and mobile telecommunications devices, includes waste stream sorting, chemical separation and treatment, decontamination, and waste logistics. Another key segment includes material recovery: metal recovery and plastic recycling contribute to both economic and environmental sustainability [1].
Waste Electrical and Electronic Equipment (WEEE) is the fastest-growing sector of solid waste, with 40-50 million tonnes generated globally each year. Only 15-20% of WEEE is recycled; the rest ends up in landfills, riverbanks and deserts, or is exported to Third World countries where it is incinerated to liberate precious metals. In the U.S. alone, 2.2 million tonnes of electronic waste (e-waste) is disposed annually [2].
The main challenge for current recycling technologies is identifying and classifying waste [3]. Machine Learning algorithms are purposefully designed for identifying and classifying data with tolerable cost...