Improvement of Waste Metal Recovery Efficiency Using Multivariate Statistical and Deep Neural Network Analysis

Abstract

In this study, we collected spectral data using laser induced breakdown spectrometry. Our results were processed by a principle component analysis (PCA) and a deep neural network to improve the separation efficiency of metals from RAM samples of a printed circuit board (PCB). The spectra were collected from 294 spots on the sample surface and subsequently divided into three groups, i.e., Black (K), Yellow (Y), and Green (G) by visual inspection of the surface color. We identified the specific wavelengths that were usable as separation criteria as well as the main elemental composition of each part by comparing the PCA and scanning electron microscopy-electron dispersive spectroscopy results. We also confirmed the possibility of automatic separation with a deep neural network whose separation accuracy was more than 98%. Our results can be used to create a new automatic separation process for waste electronic electrical equipment.