Optimization of Models for Prediction of Content of Silicon, Calcium, Magnesium and Aluminum in Refined Steel Slag Based on LIBS

Considering that industrial forming machines based on rapid analysis of the components of refined steel slag with LIBS had not been applied in the metallurgical field, models for predicting the content of silicon, calcium, magnesium and aluminum in refined steel slag were established by using the independently developed LIBS test system and analysis software. The traditional standard curve method was used for prediction of the results first, giving results of coefficient of determination (R²) of the 4 elements of 0.833 1, 0.829 4, 0.803 2 and 0.691 3, with absolute values of relative error of the predictive values in the range of 9.3%-26%. In order to optimize testing results, standard normal variate transformetion (SNV) and partial least squares (PLS) in chemometrics were used to establish the model, and the accuracy of predictive results was significantly improved. The results showed that under the condition of optimal latent variable of 15, the PLS model established after SNV pretreatment with aluminum element spectral data had a better predictive effect, with absolute value of average relative error of 6.0% and a R² of 0.950 2. The PLS model of silicon, calcium and magnesium had a good predictive effect without SNV conversion, the absolute values of relative errors of predictive values were 8.1%, 1.5% and 8.4%, and the R² were 0.928 1, 0.910 3, and 0.901 8, respectively, basically meeting the requirements of the rapid detection of steel slag.