Rapid quantitative analysis of slag acidity by laser induced breakdown spectroscopy combined with random forest

Slag is one of the industrial wastes of iron and steel smelting, the recycling of which is a research hotspot in recent years. The acidity of slag is one of the important indexes affecting the reuse of slag. Therefore, the rapid analysis of slag acidity is particularly important for industrial production and resource recovery. The feasibility of laser induced breakdown spectroscopy (LIBS) technology combined with machine learning method for the acidity analysis of slag was explored in the present work. Firstly, the LIBS spectral data of 30 slag samples were collected, and an optimal spectral pretreatment method was explored. On this basis, the variable importance measurement (VIM) based on random forest (RF) algorithm is used to screen the feature variables of LIBS spectral data of slag samples. Then, the grid search method is used to optimize the parameters of the RF calibration model. Based on the optimized spectral data and model parameters, a quantitative analysis model of slag acidity was established. In order to further verify the prediction performance of this model, it is compared with other models. The results show that the best prediction performance of slag acidity is obtained based on the combination of LIBS and VIM-RF model, of which the determination coefficient of prediction set (R2) is 0.9412, the relative analysis error (RPD) is 4.123, the root mean square error (RMSE) is 0.5358, and the average relative error (MRE) is 0.4166. This study shows that LIBS combined with VIM-RF is an effective method for rapid quantitative analysis of metallurgical slag, which can provide a reference for other index analysis in the metallurgical industry.