Rapid Predication of 8 Elements in Scrap Steel by Laser Induced Breakdown Spectroscopy with Partial Least Squares Regression

The quantitative analysis models of chromium, nickel, copper, silicon, manganese, vanadium, carbon and titanium in scrap steel were established based on laser induced breakdown spectroscopy (LIBS) with partial least squares regression (PLSR). Twelve steel standard samples were selected and scanned by a portable laser induced breakdown spectrometer for scrap steel composition. The spectral data were preprocessed with elimination, average, baseline correction and normalization. Based on the emission spectral line database of National Institute of Standards and Technology, the spectral data of modeling band of each element were selected as the model input, and the numbers of latent variable of chromium, nickel, copper, silicon, manganese, vanadium, carbon and titanium were 16, 24, 18, 22, 25, 14, 9 and 22, respectively, with the ten-fold cross-validation method. Eight elements were modeled under the optimized parameters. The results showed that the correlation coefficients of determination of the models of 8 elements were in the range of 0.957 1-0.999 6, with the root mean square errors in the range of 0.003 4-0.070 6, the average percentage errors in the range of 6.676 4-75.645, and the residual sum of squares in the range of 0.002 0-0.653 2. Except for carbon, RSD (n=5) of the determined values of the other 7 elements were not more than 7.0%. The models were used to predict manganese, silicon and copper in actual scrap steel samples, and the results were in good agreement with those obtained by spark direct reading spectrometry.