Quantitative Analysis of Carbon with Laser-Induced Breakdown Spectroscopy (LIBS) Using Genetic Algorithm and Back Propagation Neural Network Models

Abstract

Carbon content detection is an essential component of the metal-smelting and classification processes. An obstacle in carbon content detection by laser-induced breakdown spectroscopy (LIBS) of steel is the interference of carbon lines by the adjacent Fe lines. The emission line of C(I) 247.86 nm generally has higher response and transmission efficiency than the emission line of C(I) 193.09 nm, but it blends with the Fe(II) 247.86 nm line. Therefore, this study proposes a method of back propagation (BP) neural network modeling, which incorporates a genetic algorithm (GA), evaluates the method of parameter modeling and prediction based on GA to optimize the BP neural network (GA–BP), and realizes a quantitative analysis of the C(I) 247.86 nm line. The achieved root mean square error for the GA–BP model is 0.0114. The obtained linear correlation coefficient shows a significant improvement after correction, indicating that the proposed method is effective. The method is concise, easy to implement, and can be applied in the carbon content detection of steels and iron-based alloys.