Comparative Study of PLSR and SVR Using MLP Feature Extraction for Quantitative Analysis of Steel Alloy Elements by Laser-Induced Breakdown Spectroscopy

With the rapid development of the steel industry, the accurate detection of alloy element contents is of great significance for the evaluation of material properties and quality control. This study aims to establish a rapid, stable, and highly accurate quantitative detection method based on handheld LIBS to achieve effective analysis of key elements such as Fe, Cr, Mn, Ni, and Cu. To meet the demand of the steel industry for rapid, stable, and high-accuracy quantification of key alloy elements such as Cr, Mn, Ni, and Cu, this study was carried out on 20 types of standard steel spectral samples. Support Vector Regression (SVR) and Partial Least Squares Regression (PLSR) models were constructed, respectively. The SVR penalty factor C (0.1–10) and loss parameter ε (0.001–1), as well as the PLSR latent variable number Lv (1–20), were optimized using five-fold cross-validation repeated 100 times. Model performance was evaluated using the coefficient of determination (R2), root-mean-square error (RMSE), and mean relative error (MRE). In the comparison of quantitative performance, excellent predictive ability for major elements such as Fe and Cr was achieved by both models; test-set R2 values exceeded 0.92, meeting the detection requirements for high-content alloy elements. For low-content Ni, Cu, and Mn, PLSR gives R2 values of 0.92, 0.93, and 0.89, while SVR yields 0.85, 0.49, and 0.36, showing clear limitations, especially for Cu and Mn. After introducing Multilayer Perceptron feature extraction, the R2 of Ni, Cu, and Mn increases to 0.99, 0.99, and 0.97 for PLSR and to 0.99, 0.93, and 0.94 for SVR, with RMSE and MRE markedly reduced. In summary, the integration of LIBS with MLP feature extraction and PLSR offers both rapid processing capabilities and high precision, significantly improving the quantification of low-concentration elements, and is well-suited for real-time online monitoring in steel production, facilitating quality control and process optimization.