Automatic Focusing Remote Laser-Induced Breakdown Spectroscopy Analysis of Trace Elements in Steel Using Support Vector Machine Regression

Improving the automation and accuracy of remote laser-induced breakdown spectroscopy (remote-LIBS) is of great significance in the field of remote, noncontact, and in situ instrument detection. We report on a remote-LIBS device with laser auto-focusing and automatic spectral signal focusing [automatic focusing remote-LIBS (AR-LIBS)], combined with support vector regression (SVR) algorithm to enhance the system’s analytical accuracy. This system, based on a Galilean telescope (laser emission path) and a Cassegrain telescope system (plasma optical signal acquisition), enables remote noncontact component analysis within 7 m. For samples at different distances, the laser emission path automatically adjusts focus and excites plasma, while the plasma signal collection system adjusts automatically. Using 19 alloy steel samples as the training set to establish a calibration model and seven samples as the validation set for verification. In order to improve the analysis performance of LIBS, four models for determining the content of Mn, Ni, Cr, and Si elements in alloy steel samples were established based on feature peak intensity as input, using univariate regression (UR) and SVR. The linear fitting degree ( $R^{2}$ ), root-mean-square error of calibration (RMSEC), root-mean-square error of prediction (RMSEP), mean absolute percentage error (MAPE), and relative standard deviation (RSD) of Mn, Ni, Cr, and Si elements in different quantitative models were evaluated. The results show that the SVR model based on the internal standard method-SVR (IS-SVR) has higher accuracy and precision. The $R^{2}$ of the Mn, Ni, Cr, and Si elements training sets is all above 0.95, with predicted MAPEs of 4.49%, 5.19%, 4.59%, and 11.54%, and predicted RSDs of 3.26%, 7.58%, 3.09%, and 3.06%, respectively.