Microstructure classification of steel samples with different heat-treatment processes based on laser-induced breakdown spectroscopy (LIBS)

Abstract

This study explores the application of laser-induced breakdown spectroscopy (LIBS) to classify steel samples, which gives a novel idea of utilizing the matrix effect. In engineering applications, carbon-steel, which has the same elemental composition, is usually processed into different microstructures through heat treatment processes. It results in the steel having different element distribution characteristics at the microscopic scale, and is considered to be one of the reasons for the matrix effect in the LIBS field. In this study, the matrix effect of LIBS spectra is used as the feature for microstructure classification of carbon-steel. According to this idea, our study introduces a rapid classification method of LIBS spectra using the random projection (RP) technique in convolutional neural networks (CNNs), which has achieved the accuracy of 99% in 25 seconds. The experimental results show that the dimensionality reduction without spectral preprocessing by the RP-CNN method enhances the impact of matrix effect. This study provides an efficient deep learning method for similar LIBS spectra obtained from steel samples with different microstructures, which has great potential in the LIBS application of engineering material evaluation.